1. The Motivation and Story Behind It
As electronic engineer and audiophile, I've always dreamed of building an amplifier that would satisfy my own standards. This would not only test my professional knowledge but also be a way to combine my hobby with my work. Though the idea sounds wonderful, the reality was far from easy. Constant interruptions and delays kept pushing back this grand plan.In truth, the delays weren't just because I was busy, but also because I was unsure of what kind of amplifier I wanted to DIY. Common designs on the market—whether tube, BJT, or MOSFET—didn't seem to be the answer I was looking for. After years of indecision, I finally made a choice in the summer of 2020: to build a no-feedback amplifier that would meet the standards of high-end amplifiers. When I made this decision, it even shocked me. As an electronics engineer, my instinct told me it was an almost impossible task. Despite countless thoughts of giving up, my competitive spirit kept bringing me back. I was too curious to know what a no-feedback amplifier would sound like. This question lingered in my mind for a long time. Out of curiosity, I asked ChatGPT, and here's the response I got (Oh, how I wish ChatGPT had been around back then):
Thus began my DIY journey, filled with challenges and joys, to experience the sound of a fully no-feedback amplifier.
Since Bell Labs engineer Harold S. Black introduced the concept of negative feedback in 1927, it quickly gained widespread application in audio amplifiers, achieving great success. For example, the Williamson feedback high-fidelity amplifier, released in 1947, used deep global feedback to significantly reduce distortion, extend frequency response, lower output impedance, and increase damping factor. These improvements made the Williamson amplifier the benchmark for high-fidelity sound systems at the time and a classic cherished by audiophiles today. Most modern amplifiers are still designed with appropriate feedback to improve measurable data.
However, humanity never lacks pioneers who dare to challenge conventions. Many engineers have been working to reduce the amount of feedback, hoping to achieve better sound through shallow feedback. Products from brands like First Watt, Pass Labs, and NAT Audio are successful examples of this approach. These amplifiers remove global feedback and only use local feedback to maintain performance, aiming to reduce the negative impact of global feedback on sound quality. Though their test data may not be perfect, they focus on sound performance and have been well-received in the market.
The success of these products greatly inspired me. I believe that following in the footsteps of earlier engineers can take me further. As an engineer, I value objective data, but as an audiophile, I also think subjective listening experience is crucial. After all, good test data doesn’t always guarantee good sound, and if the sound is pleasant, why worry about perfect data? The quality of an amplifier cannot be entirely defined by one set of parameters, at least not with today’s technology. Objective data gives engineers direction, but subjective listening remains the ultimate measure of an amplifier's quality.
So why did I ultimately choose to challenge myself with the “hell-level” difficulty of a no-feedback amplifier? There’s a small story here. I had a friend who was an audiophile. He once told me that he had listened to amplifiers without global feedback and found the sound more natural, lively, and impactful. He asked if I could build a fully no-feedback amplifier for him to try. At the time, I just smiled and didn’t give it much thought. But his words planted a seed in my mind, which eventually led me to make this decision and explore the mystery myself.
In fact, I wasn’t sure what a no-feedback amplifier would sound like either, or if it would really be as natural and lively as ChatGPT suggested. But debates in technology often have their complexities. Feedback indeed brings many benefits but can also introduce issues like transient intermodulation distortion and dynamic errors caused by feedback delay. Could these distortions negatively affect subjective listening? Perhaps building a no-feedback amplifier with excellent parameters is the best way to answer these questions.
This is the story and motivation behind my DIY no-feedback amplifier, as well as my views on audio technology and subjective evaluation. In the upcoming posts, I’ll share why I chose Class D as the main architecture, and the challenges and results I encountered during simulation, production, and debugging. This DIY journey took four years, and some details may have faded from memory, but I’ll try to consider readers of all backgrounds, using analogies and simplified technical details to help everyone understand. I hope my sharing will bring you some inspiration.
Last edited:
Perhaps you should have read Bruno Putzey's article about feedback....
https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf
https://linearaudio.net/sites/linearaudio.net/files/volume1bp.pdf
Open loop class D is old hat.
Examples of stuff we did many years ago include TI TAS5508C / TAS5162 and ST STA309A / STA508.
PSRR is non-existent in these types of amplifiers.
Examples of stuff we did many years ago include TI TAS5508C / TAS5162 and ST STA309A / STA508.
PSRR is non-existent in these types of amplifiers.
...and a regulated supply would be unacceptable, because voltage regulators use feedback. I'm curious how flmhhh has solved this: batteries, stacks of supercapacitors?
Based on feedback from friends about "The Journey of DIYing a No-Feedback Class D Amplifier (1)," I need to clarify a few points:
After deciding to build a no-feedback amplifier, the next step was to determine which amplifier architecture to adopt. Looking back now, the choice seems straightforward, but at the time, it involved a lot of back-and-forth deliberation.
For solid-state devices like transistors or MOSFETs, whether in Class A, Class AB, or Class G designs, removing global negative feedback is relatively easy. However, these architectures’ current output stages often inherently feature voltage follower circuits, which naturally incorporate feedback, contradicting the goal of having no feedback at all. As a result, these options were quickly ruled out.
I had also considered using vacuum tube circuits, but as an engineer raised in the transistor era, the thought of using vacuum tubes made me uneasy. Additionally, the difficulty in producing high-quality output transformers for tube amps, where even minor issues could severely affect sound quality, led me to abandon this idea.
After evaluating these factors, Class D emerged as the only feasible option. Class D amplifiers can provide stable output with decent sound quality, even without audio feedback, and maintain a THD+N in the range of 0.1% to 1%. Figure 1 shows a simplified diagram of a Class D amplifier. From the small-signal input to the high-current output stage, there is no feedback loop. The appeal of Class D lies in its potential to achieve distortion-free output, at least theoretically, when components are close to ideal conditions.
Figure 1， An Example of Analog Class D Amplifier
(from ‘Audio Power Amplifiers’ By Bob Cordell)
At that time, there were also many high-performance Class D amplifiers on the market. These products significantly reduced distortion by incorporating audio feedback techniques specifically designed for Class D amplifiers. It must be acknowledged that feedback is indeed an effective means of reducing static distortion. Figure 2 shows a basic schematic of a Class D amplifier with audio feedback.
Figure 2, An Example of Analog Class D Amplifier with audio negative feedback
(from ‘Comprehensive Study of Class D Amplifier Technology’ by Todd P.Marco)
Certainly, Class D amplifiers come in many variations. The two main types are fixed-frequency (PWM modulation, shown in Figure 3(a)) and self-oscillating Class D (PDM modulation, shown in Figure 3(b)) amplifiers. Additionally, there are two main designs for the output stage: half-bridge (shown in Figure 4(a)) and full-bridge (shown in Figure 4(b)). Within the full-bridge design, there are further distinctions, such as AD modulation and BD modulation.
(a) (b)
I don’t plan to go into detailed explanations of the various technologies here, as there is already a lot of papers and books that thoroughly explore these modulation methods. What I want to convey is that even after choosing the Class D amplifier architecture, it wasn’t as straightforward as imagined. In addition to reviewing a large amount of literature, further exploration was needed because most research includes feedback functions, and there are almost no readily available, low-distortion, no-feedback design examples to refer to. Nevertheless, I feel that regardless of the outcome, I should proceed with designing and building this amplifier. Even if it ultimately fails, I have no regrets, as I have invested a great deal of passion and effort up to this point, and my dream continues to drive me forward.
Next comes the overall design scheme for the Class D amplifier. For clarity, I will list the carefully considered overall scheme and choices in the table below, with explanations for the different technical choices to be provided during subsequent simulation and debugging. In fact, these schemes were not determined from the beginning but were gradually refined through simulation, construction, and debugging.
I chose to design a headphone amplifier due to its relatively low output power, allowing for the use of smaller power MOSFETs. Typically, small power MOSFETs have lower gate capacitance and can operate at higher frequencies. This increase in frequency not only significantly reduces the amplifier's distortion but also simplifies the design of the output filter. Additionally, a higher cutoff frequency means that variations in the filter's Q factor have a minimal impact on the amplifier’s frequency response when connecting headphones with different impedances.
Next, I will share some key ideas of circuit design ideas and overall simulation results, as well as discuss specific measures for reducing distortion. Stay tuned.
- "No-feedback" means that neither global nor local audio signal feedback is used in the amplifier design. However, to ensure circuit stability, other functional circuits may still include feedback mechanisms, both positive and negative.
- I have no bias against feedback. On the contrary, I understand that feedback is extremely successful and indispensable in many applications.
- The goal of this DIY project is purely out of curiosity and a desire to challenge myself. I want to see how well I can control distortion in an amplifier without using audio signal feedback, and to discover what unique characteristics the sound might have under such conditions.
After deciding to build a no-feedback amplifier, the next step was to determine which amplifier architecture to adopt. Looking back now, the choice seems straightforward, but at the time, it involved a lot of back-and-forth deliberation.
For solid-state devices like transistors or MOSFETs, whether in Class A, Class AB, or Class G designs, removing global negative feedback is relatively easy. However, these architectures’ current output stages often inherently feature voltage follower circuits, which naturally incorporate feedback, contradicting the goal of having no feedback at all. As a result, these options were quickly ruled out.
I had also considered using vacuum tube circuits, but as an engineer raised in the transistor era, the thought of using vacuum tubes made me uneasy. Additionally, the difficulty in producing high-quality output transformers for tube amps, where even minor issues could severely affect sound quality, led me to abandon this idea.
After evaluating these factors, Class D emerged as the only feasible option. Class D amplifiers can provide stable output with decent sound quality, even without audio feedback, and maintain a THD+N in the range of 0.1% to 1%. Figure 1 shows a simplified diagram of a Class D amplifier. From the small-signal input to the high-current output stage, there is no feedback loop. The appeal of Class D lies in its potential to achieve distortion-free output, at least theoretically, when components are close to ideal conditions.
Figure 1， An Example of Analog Class D Amplifier
(from ‘Audio Power Amplifiers’ By Bob Cordell)
At that time, there were also many high-performance Class D amplifiers on the market. These products significantly reduced distortion by incorporating audio feedback techniques specifically designed for Class D amplifiers. It must be acknowledged that feedback is indeed an effective means of reducing static distortion. Figure 2 shows a basic schematic of a Class D amplifier with audio feedback.
Figure 2, An Example of Analog Class D Amplifier with audio negative feedback
(from ‘Comprehensive Study of Class D Amplifier Technology’ by Todd P.Marco)
Certainly, Class D amplifiers come in many variations. The two main types are fixed-frequency (PWM modulation, shown in Figure 3(a)) and self-oscillating Class D (PDM modulation, shown in Figure 3(b)) amplifiers. Additionally, there are two main designs for the output stage: half-bridge (shown in Figure 4(a)) and full-bridge (shown in Figure 4(b)). Within the full-bridge design, there are further distinctions, such as AD modulation and BD modulation.
(a) (b)
Figure 3, PWM waveform (a) and PDM waveform (b) of class D amplifier
(from ‘Comprehensive Study of Class D Amplifier Technology’ by Todd P.Marco)
Figure 4，Half bridge (a) and full bridge (b) class D output stages
(from ‘Audio Power Amplifiers’ By Bob Cordell)
(from ‘Comprehensive Study of Class D Amplifier Technology’ by Todd P.Marco)
Figure 4，Half bridge (a) and full bridge (b) class D output stages
(from ‘Audio Power Amplifiers’ By Bob Cordell)
I don’t plan to go into detailed explanations of the various technologies here, as there is already a lot of papers and books that thoroughly explore these modulation methods. What I want to convey is that even after choosing the Class D amplifier architecture, it wasn’t as straightforward as imagined. In addition to reviewing a large amount of literature, further exploration was needed because most research includes feedback functions, and there are almost no readily available, low-distortion, no-feedback design examples to refer to. Nevertheless, I feel that regardless of the outcome, I should proceed with designing and building this amplifier. Even if it ultimately fails, I have no regrets, as I have invested a great deal of passion and effort up to this point, and my dream continues to drive me forward.
Next comes the overall design scheme for the Class D amplifier. For clarity, I will list the carefully considered overall scheme and choices in the table below, with explanations for the different technical choices to be provided during subsequent simulation and debugging. In fact, these schemes were not determined from the beginning but were gradually refined through simulation, construction, and debugging.
Class D Overall Design Scheme and Requirements | |
| Amplifier Type | Headphone Amplifier |
| Frequency Response | 5Hz~30kHz |
| Max output power | 3000mW |
| Modulation Method | Sigima-Delta (PDM) |
| Modulation Frequency | >2MHz |
| Integration Order | Multiple stages |
| Integration Circuit | Differential Integration |
| Output stage | Full Bridge |
| Output Filter | Three-Level LC Filter |
| Load Impedance | 32 ~ 600 Ohm |
| Power Supply | Lithium Battery |
| Power Supply Circuit | DC/DC + Class A, DC-DC + LDO |
I chose to design a headphone amplifier due to its relatively low output power, allowing for the use of smaller power MOSFETs. Typically, small power MOSFETs have lower gate capacitance and can operate at higher frequencies. This increase in frequency not only significantly reduces the amplifier's distortion but also simplifies the design of the output filter. Additionally, a higher cutoff frequency means that variations in the filter's Q factor have a minimal impact on the amplifier’s frequency response when connecting headphones with different impedances.
Next, I will share some key ideas of circuit design ideas and overall simulation results, as well as discuss specific measures for reducing distortion. Stay tuned.
Attachments
Last edited:
You could probably simplify life just a little by starting with just a single switch. A half-bridge optimizes power (etc.) and a full-bridge optimizes things even more, but they both present unique challenges like deadtime to prevent shoot-through, to name one. Don't get bogged down.
If you were me (and for whatever reason I was set on class-D), I would start with the basics: a nice triangle wave, a nice constant current source and LPF to charge linearly. In addition, I would try to find ways to test the "lower" limits. Can I use a 40kHz oscillator? What do I need to improve to make it sound good? First make a PC speaker sound like a Soundblaster, then ramp up the frequency.
If you were me (and for whatever reason I was set on class-D), I would start with the basics: a nice triangle wave, a nice constant current source and LPF to charge linearly. In addition, I would try to find ways to test the "lower" limits. Can I use a 40kHz oscillator? What do I need to improve to make it sound good? First make a PC speaker sound like a Soundblaster, then ramp up the frequency.
Thank you for the great suggestion, much appreciated!
The entire DIY project has been completed after about a four-year journey. Now, I’m in the process of revisiting everything, documenting it, and sharing the experience.
Of course, I will continue to refine and improve the design as I receive more valuable suggestions or ideas.
The entire DIY project has been completed after about a four-year journey. Now, I’m in the process of revisiting everything, documenting it, and sharing the experience.
Of course, I will continue to refine and improve the design as I receive more valuable suggestions or ideas.
Perhaps you missed this key sentence in your attachment:
"In contrast, ultrasound, whose frequency ranges up to at least 120 kHz, can create an auditory sensation when
delivered via bone conduction (BC) [2– 4]. "
"In contrast, ultrasound, whose frequency ranges up to at least 120 kHz, can create an auditory sensation when
delivered via bone conduction (BC) [2– 4]. "
3. Simulation Results and Reflections on the Circuit Design
After finalizing the overall design, simulations are used to select specific components and determine resistor and capacitor values, followed by circuit board manufacturing for validation and iteration to achieve the initial design goals.
The Σ-Δ (Sigma-Delta) modulation method was chosen for the following reasons:
1，A fixed-frequency PWM modulator has almost zero Power Supply Rejection Ratio(PSRR) in its output stage, meaning it has no ability to suppress power noise. This creates significant challenges for power supply design, as any noise from the power supply directly converts into audio noise and reaches the headphones. In contrast, the Σ-Δ modulator offers some level of power supply noise rejection, greatly reducing the burden on power supply design.
2，Fixed-frequency PWM modulators use a standard triangular wave (blue waveform in Figure 1(a)), which contains a lot of high-frequency components. However, the integrator-generated waveforms from multi-stage integrators (green waveform in Figure 1(a)) resemble a triangular shape but with fewer high-frequency components and faster high-frequency roll-off (as shown in Figure 1(b), with blue representing the standard triangular wave's spectrum and green representing the multi-stage integrator waveform's spectrum). In an amplifier with the same bandwidth, the multi-stage integrator waveform results in lower distortion, and the PDM modulation signal's edge jitter (rise and fall times) is smaller, leading to overall lower noise and distortion of audio signal.
(a) (b)
Figure 1, Triangle Wave, Higher-Order Integral Wave, and Their Spectrum
3，The Σ-Δ modulator can use a differential integrator circuit, which outputs a pair of equal-amplitude, opposite-polarity integrator waveforms (see Figure 2(b)) to the comparator. In contrast, the fixed-frequency PWM modulator sends a triangular wave and input signal to the comparator (Figure 2(a) shows the comparator input waveform of the fixed-frequency PWM modulator). It's clear that the slope (rate of change) at the intersection points of these two sets of input waveforms differs significantly. The differential integrator waveforms have a much steeper slope at the intersection points compared to the PWM modulator’s waveforms. This means that for the same noise level, the edge jitter of the comparator output is smaller, thus improving the overall amplifier performance. Attentive readers might have figured out that the slope at the signal crossover points of the higher-order integrator waveform should be much steeper than that of the standard triangular wave, which could be one of the reasons why the higher-order integrator performs better than the standard triangular wave.
(a) (b)
Figure 2， Comparator Input Signals of PWM and PDM Modulators
4，Compared to fixed-frequency PWM modulators, Σ-Δ modulators offer greater flexibility and room for improvement in design. For example, a Schmitt trigger mechanism can be added to the comparator in a Σ-Δ modulator to improve the stability of the PDM pulse waveform edges.
The discussion so far has focused on reducing edge jitter in the PDM pulse signal. In Class D amplifiers, the pulse signal is central to the entire system; only precise pulses can result in low noise and low THD distortion. The edges of these pulses are crucial. If edge jitter is unrelated to the input signal, it generates noise, which, while having a smaller effect on distortion, can increase the amplifier’s background noise. If edge jitter is related to the input signal—such as when the triangular wave lacks linearity—the jitter (early or late edges) will correlate with the input signal amplitude, causing nonlinear distortion and resulting in higher THD.
Through the improvements discussed above, Σ-Δ modulation can potentially achieve lower distortion. The next step is to compare performance through simulation to select different components and parameters. This is a complex and tedious process, as components interact with each other, and changes to one typically require adjustments to related parameters. I'll use a key example to explain this further.
The precision of the modulated pulse signal comes from the accuracy of the multi-order integrator signals. The amplitude and frequency of these integrator waveforms are affected by many factors, such as the zero and pole distribution in the integrator network, the depth of the Schmitt trigger in the comparator, and the oscillation control resistor. For simplicity, Figure 3 shows a basic schematic of a multi-order half-bridge Σ-Δ modulator with a Schmitt trigger. The ratio of the Schmitt trigger resistor influences the modulation frequency, and the equivalent capacitance of the high-order integrator also affects the modulation frequency. These factors further shape the integrator waveform and its amplitude.
The accuracy of the integrator waveform has already been analyzed, but its amplitude is equally important, requiring a balance between the output capability of the integrator and the performance of the comparator. In general, a larger signal fed into the comparator results in more precise outputs, but an overly large output from the integrator can cause signal distortion.
Figure 3， Multistage Half-Bridge Sigma-Delta Modulator with Schmitt Trigger Mechanism
The second important aspect is the selection of the output stage, where the industry has reached a consensus: when cost requirements are not stringent, a full-bridge configuration is typically used. The advantages of the full-bridge mode include the ability to partially cancel even-order harmonics and suppress the back current of the output filter inductor, effectively reducing the power supply fluctuations in the output stage. This is particularly beneficial for Class D amplifiers with poor power supply rejection ratio (PSRR).
The output filter is also critical, especially for Class D amplifiers using PDM modulation. Since this design uses a modulation frequency above 2 MHz, there is more flexibility in choosing the low-pass filter’s cutoff frequency. This brings significant benefits: first, a higher cutoff frequency can be chosen, so the frequency response changes with different loads won't significantly impact the amplifier’s effective bandwidth. Second, since smaller inductors are used, they provide better linearity at the same size, reducing the distortion impact on the amplifier. During simulation, it was also observed that a multi-order output filter could reduce in-band noise, although there is no theoretical explanation for this yet. I welcome any interested parties to discuss this further.
The power supply solution is equally important. This design uses a single 3000mAh lithium battery, so a step-up boost converter is required. A major issue is that the amplifier operates at over 2 MHz, which is much higher than the frequency of commonly available boost converters. During a single PWM cycle, the supply voltage may drop due to the amplifier’s high current draw, which even the best boost chips struggle to handle in such cases.
After multiple experiments, both the pre-amp and output stages ended up using a two-stage single power supply solution. The pre-amp power (for the integrator and comparator) uses a combination of a boost circuit and a Class A power supply. The Class A supply uses a high-precision reference regulator with voltage accuracy of 0.1% and power supply noise as low as a few microvolts. Additionally, the midpoint reference voltage uses the same precision reference chip, not only improving voltage accuracy but also ensuring the integrator circuit operates symmetrically, optimizing open-loop performance.
Originally, I planned to use the same design for the output stage power supply, but due to the need to drive headphones with varying impedance and sensitivity, the Class A power supply’s efficiency was too low, which affected battery life. In the end, I chose a step-up circuit combined with a high-precision, high-current LDO and large capacitors for the output stage.
Interestingly, switching from a Class A supply to an LDO + large capacitors for the output stage resulted in virtually no subjective change in sound quality. However, changing the pre-amp from an LDO to a Class A power supply led to a significant improvement in sound. The high frequencies became more transparent, the low frequencies were stronger and more extended, and the soundstage was more stable. This change was quite noticeable, without the need for careful listening. Could this be due to the higher precision in both supply and bias voltages? I would love to hear thoughts from those interested in discussing this. The first version of the pre-amp, which was discarded, used a voltage divider for the bias reference, as shown in Figure 4, with resistor accuracy of 0.1%.
Figure 4， Circuit for Generating the Bias Voltage in the Integrator
There are many other crucial details in the amplifier design and debugging process that cannot be overlooked, and I will discuss them further in appropriate contexts.
Figure 5 shows the Spice simulation results for the background noise. Within an effective bandwidth of 35 kHz, the highest background noise is as low as -170 dBV, with the low-frequency section reaching nearly -180 dBV.
Figure 5，Zero Input Circuit Configuration (Background Noise) Simulation Results
Figure 6 shows the simulation results with a 1kHz input signal. No harmonic distortion related to the 1kHz frequency was observed. However, compared to the zero-input results in Figure 5, the background noise increased by about 10dB, reaching approximately -170dBV. This frequency-independent distortion is a typical occurrence in such circuits.
Figure 6， Single Sine Wave Input Circuit Configuration Simulation Results
Figure 7 presents the amplifier simulation results based on the previous discussions. The modulation frequency is around 2.3 MHz, with an input signal peak-to-peak value of about 196 mV and an output peak-to-peak value of around 4000 mV, resulting in an overall gain of approximately 26 dB. The input signal consists of three frequencies: 500 Hz, 5 kHz, and 20 kHz. The amplifier's background noise level is about -167 dBV, which is 3 dB higher than the -170 dBV observed in the 1 kHz sine wave test shown in Figure 6, likely due to frequency-independent distortion causing this noise floor increase.
During the simulation, it was observed that different multi-stage integrator network configurations affect the spectral characteristics of frequency-independent distortion. By adjusting the integrator network, it is possible to reduce low-frequency distortion at the cost of increasing high-frequency distortion. In this design, I chose to balance the frequency-independent distortion across the spectrum, leading to a flat noise floor.
In reality, there are other approaches. One could adjust the integrator network to minimize the total or weighted power of frequency-independent noise within the effective bandwidth, without prioritizing flatness. My choice for a flat noise floor was based on the fact that amplifiers typically exhibit more distortion in high frequencies compared to low frequencies, whether it is frequency-dependent or independent. A DIY amplifier with a flat noise floor might offer a different subjective listening experience.
Given the multi-frequency input, the amplifier's total distortion is primarily composed of harmonic and intermodulation distortion. The simulation results show that the highest harmonic distortion reaches about -156 dBV, with the minimum signal-to-harmonic ratio being 152 dB. These results demonstrate excellent noise suppression and distortion control, achieving the expected high-performance standards, at least based on the simulation outcomes.
Figure 7，Multi-Sine Wave Input Circuit Configuration Simulation Results
The following lists the different distortion frequencies and their amplitudes generated from the simulation, for reference by those interested:
For those who might not be familiar with these simulation figures and find them hard to interpret, it's worth noting that if you were to input a true 24-bit audio source into the amplifier, the output would be virtually undistorted. The harmonic distortion is more than 7 dB lower than the 24-bit quantization noise, making it effectively distortion-free.
In addition, I added negative feedback to the circuit from Figure 7 and adjusted the overall gain to 20 dB, 6 dB lower than without feedback. Observing the results in Figure 8, no frequency-dependent harmonics were observed, but the overall noise floor increased by about 1 dB compared to the no-feedback case.
This phenomenon may differ from traditional Class B amplifiers. Class D amplifiers, due to their modulation frequency, generate out-of-band noise. When negative feedback is introduced, this out-of-band noise can also be fed back to the input. While negative feedback effectively suppresses harmonic distortion caused by nonlinearity, the feedback of out-of-band noise may increase frequency-independent distortion, thus raising the overall noise floor.
Although this analysis partially explains the rise in noise floor, it has not been rigorously validated and is provided for reference only. If I gather further experimental data or validation, I will share it separately in the future.
Figure 8， Simulation Results of Multi-Sine Wave Input Circuit Scheme with Global Feedback
Additionally, when observing the operating frequency of the amplifier, it becomes apparent that as the peak value of the input signal increases, the divergence of the main frequency enhances, while the peak itself becomes lower. This phenomenon is a characteristic of the sigma-delta self-oscillation architecture, providing significant advantages in electromagnetic compatibility (EMC). Since the self-oscillation frequency is not fixed, it dynamically changes under different load and signal conditions, dispersing EMI (electromagnetic interference) energy rather than concentrating it at a specific frequency. This reduces interference with other devices, making sigma-delta amplifiers relatively superior in EMC performance in practical applications.
The following text is excerpted from "Audio Power AMP Design Handbook 4th edition" by Douglas Self, which discusses the increasing distortion characteristics of traditional Class B amplifiers at higher frequencies. Although this book was published in 2006, and advancements in technology may have significantly improved the high-frequency distortion characteristics of current amplifiers, this fundamental principle still holds unless there are breakthrough developments.
From the simulation results, Class D amplifiers also adhere to this pattern, but their high-frequency distortion is only a few dB higher than the low-frequency distortion. Given the overall low distortion rates, this difference can generally be ignored.
Another noteworthy aspect is that the amplifier's phase response tends to be linear. As we know, the human ear is not very sensitive to phase response in sound. Therefore, it raises an interesting question: do linear-phase amplifiers produce different subjective listening experiences? If anyone has personal experiences or insights on this, it would be great to share and discuss!
Based on a simple calculation of the simulated distortion effects, the amplifier's THD value can reach approximately 0.0000025% with the minimal SNR is 152dB. Considering that the harmonic components are several dB above the noise floor, the THD+N value should be in the same range. This simulation result seems almost unrealistically good, so I conducted multiple verification simulations and cross-checked with actual test results, ultimately confirming that the simulation results have a certain degree of credibility. The verification approach was as follows:
Of course, the performance results from the simulation represent only a theoretical upper limit; the actual amplifier's performance may degrade to varying degrees due to various factors. A good simulation result is merely a starting point, and the process of building the amplifier is essentially a struggle against these degradation factors.
Writing the threads has been particularly challenging for me. I’ve tried to clearly describe the simulation conditions at the time, but the reality is far more complex than what I've expressed here. If there are any unclear or incorrect points, I welcome further discussion. For now, I will share the results of my two major version of DIY AMP and some subjective listening experiences, so stay tuned!
After finalizing the overall design, simulations are used to select specific components and determine resistor and capacitor values, followed by circuit board manufacturing for validation and iteration to achieve the initial design goals.
The Σ-Δ (Sigma-Delta) modulation method was chosen for the following reasons:
1，A fixed-frequency PWM modulator has almost zero Power Supply Rejection Ratio(PSRR) in its output stage, meaning it has no ability to suppress power noise. This creates significant challenges for power supply design, as any noise from the power supply directly converts into audio noise and reaches the headphones. In contrast, the Σ-Δ modulator offers some level of power supply noise rejection, greatly reducing the burden on power supply design.
2，Fixed-frequency PWM modulators use a standard triangular wave (blue waveform in Figure 1(a)), which contains a lot of high-frequency components. However, the integrator-generated waveforms from multi-stage integrators (green waveform in Figure 1(a)) resemble a triangular shape but with fewer high-frequency components and faster high-frequency roll-off (as shown in Figure 1(b), with blue representing the standard triangular wave's spectrum and green representing the multi-stage integrator waveform's spectrum). In an amplifier with the same bandwidth, the multi-stage integrator waveform results in lower distortion, and the PDM modulation signal's edge jitter (rise and fall times) is smaller, leading to overall lower noise and distortion of audio signal.
(a) (b)
Figure 1, Triangle Wave, Higher-Order Integral Wave, and Their Spectrum
3，The Σ-Δ modulator can use a differential integrator circuit, which outputs a pair of equal-amplitude, opposite-polarity integrator waveforms (see Figure 2(b)) to the comparator. In contrast, the fixed-frequency PWM modulator sends a triangular wave and input signal to the comparator (Figure 2(a) shows the comparator input waveform of the fixed-frequency PWM modulator). It's clear that the slope (rate of change) at the intersection points of these two sets of input waveforms differs significantly. The differential integrator waveforms have a much steeper slope at the intersection points compared to the PWM modulator’s waveforms. This means that for the same noise level, the edge jitter of the comparator output is smaller, thus improving the overall amplifier performance. Attentive readers might have figured out that the slope at the signal crossover points of the higher-order integrator waveform should be much steeper than that of the standard triangular wave, which could be one of the reasons why the higher-order integrator performs better than the standard triangular wave.
(a) (b)
Figure 2， Comparator Input Signals of PWM and PDM Modulators
4，Compared to fixed-frequency PWM modulators, Σ-Δ modulators offer greater flexibility and room for improvement in design. For example, a Schmitt trigger mechanism can be added to the comparator in a Σ-Δ modulator to improve the stability of the PDM pulse waveform edges.
The discussion so far has focused on reducing edge jitter in the PDM pulse signal. In Class D amplifiers, the pulse signal is central to the entire system; only precise pulses can result in low noise and low THD distortion. The edges of these pulses are crucial. If edge jitter is unrelated to the input signal, it generates noise, which, while having a smaller effect on distortion, can increase the amplifier’s background noise. If edge jitter is related to the input signal—such as when the triangular wave lacks linearity—the jitter (early or late edges) will correlate with the input signal amplitude, causing nonlinear distortion and resulting in higher THD.
Through the improvements discussed above, Σ-Δ modulation can potentially achieve lower distortion. The next step is to compare performance through simulation to select different components and parameters. This is a complex and tedious process, as components interact with each other, and changes to one typically require adjustments to related parameters. I'll use a key example to explain this further.
The precision of the modulated pulse signal comes from the accuracy of the multi-order integrator signals. The amplitude and frequency of these integrator waveforms are affected by many factors, such as the zero and pole distribution in the integrator network, the depth of the Schmitt trigger in the comparator, and the oscillation control resistor. For simplicity, Figure 3 shows a basic schematic of a multi-order half-bridge Σ-Δ modulator with a Schmitt trigger. The ratio of the Schmitt trigger resistor influences the modulation frequency, and the equivalent capacitance of the high-order integrator also affects the modulation frequency. These factors further shape the integrator waveform and its amplitude.
The accuracy of the integrator waveform has already been analyzed, but its amplitude is equally important, requiring a balance between the output capability of the integrator and the performance of the comparator. In general, a larger signal fed into the comparator results in more precise outputs, but an overly large output from the integrator can cause signal distortion.
Figure 3， Multistage Half-Bridge Sigma-Delta Modulator with Schmitt Trigger Mechanism
The second important aspect is the selection of the output stage, where the industry has reached a consensus: when cost requirements are not stringent, a full-bridge configuration is typically used. The advantages of the full-bridge mode include the ability to partially cancel even-order harmonics and suppress the back current of the output filter inductor, effectively reducing the power supply fluctuations in the output stage. This is particularly beneficial for Class D amplifiers with poor power supply rejection ratio (PSRR).
The output filter is also critical, especially for Class D amplifiers using PDM modulation. Since this design uses a modulation frequency above 2 MHz, there is more flexibility in choosing the low-pass filter’s cutoff frequency. This brings significant benefits: first, a higher cutoff frequency can be chosen, so the frequency response changes with different loads won't significantly impact the amplifier’s effective bandwidth. Second, since smaller inductors are used, they provide better linearity at the same size, reducing the distortion impact on the amplifier. During simulation, it was also observed that a multi-order output filter could reduce in-band noise, although there is no theoretical explanation for this yet. I welcome any interested parties to discuss this further.
The power supply solution is equally important. This design uses a single 3000mAh lithium battery, so a step-up boost converter is required. A major issue is that the amplifier operates at over 2 MHz, which is much higher than the frequency of commonly available boost converters. During a single PWM cycle, the supply voltage may drop due to the amplifier’s high current draw, which even the best boost chips struggle to handle in such cases.
After multiple experiments, both the pre-amp and output stages ended up using a two-stage single power supply solution. The pre-amp power (for the integrator and comparator) uses a combination of a boost circuit and a Class A power supply. The Class A supply uses a high-precision reference regulator with voltage accuracy of 0.1% and power supply noise as low as a few microvolts. Additionally, the midpoint reference voltage uses the same precision reference chip, not only improving voltage accuracy but also ensuring the integrator circuit operates symmetrically, optimizing open-loop performance.
Originally, I planned to use the same design for the output stage power supply, but due to the need to drive headphones with varying impedance and sensitivity, the Class A power supply’s efficiency was too low, which affected battery life. In the end, I chose a step-up circuit combined with a high-precision, high-current LDO and large capacitors for the output stage.
Interestingly, switching from a Class A supply to an LDO + large capacitors for the output stage resulted in virtually no subjective change in sound quality. However, changing the pre-amp from an LDO to a Class A power supply led to a significant improvement in sound. The high frequencies became more transparent, the low frequencies were stronger and more extended, and the soundstage was more stable. This change was quite noticeable, without the need for careful listening. Could this be due to the higher precision in both supply and bias voltages? I would love to hear thoughts from those interested in discussing this. The first version of the pre-amp, which was discarded, used a voltage divider for the bias reference, as shown in Figure 4, with resistor accuracy of 0.1%.
Figure 4， Circuit for Generating the Bias Voltage in the Integrator
There are many other crucial details in the amplifier design and debugging process that cannot be overlooked, and I will discuss them further in appropriate contexts.
Figure 5 shows the Spice simulation results for the background noise. Within an effective bandwidth of 35 kHz, the highest background noise is as low as -170 dBV, with the low-frequency section reaching nearly -180 dBV.
Figure 5，Zero Input Circuit Configuration (Background Noise) Simulation Results
Figure 6 shows the simulation results with a 1kHz input signal. No harmonic distortion related to the 1kHz frequency was observed. However, compared to the zero-input results in Figure 5, the background noise increased by about 10dB, reaching approximately -170dBV. This frequency-independent distortion is a typical occurrence in such circuits.
Figure 6， Single Sine Wave Input Circuit Configuration Simulation Results
Figure 7 presents the amplifier simulation results based on the previous discussions. The modulation frequency is around 2.3 MHz, with an input signal peak-to-peak value of about 196 mV and an output peak-to-peak value of around 4000 mV, resulting in an overall gain of approximately 26 dB. The input signal consists of three frequencies: 500 Hz, 5 kHz, and 20 kHz. The amplifier's background noise level is about -167 dBV, which is 3 dB higher than the -170 dBV observed in the 1 kHz sine wave test shown in Figure 6, likely due to frequency-independent distortion causing this noise floor increase.
During the simulation, it was observed that different multi-stage integrator network configurations affect the spectral characteristics of frequency-independent distortion. By adjusting the integrator network, it is possible to reduce low-frequency distortion at the cost of increasing high-frequency distortion. In this design, I chose to balance the frequency-independent distortion across the spectrum, leading to a flat noise floor.
In reality, there are other approaches. One could adjust the integrator network to minimize the total or weighted power of frequency-independent noise within the effective bandwidth, without prioritizing flatness. My choice for a flat noise floor was based on the fact that amplifiers typically exhibit more distortion in high frequencies compared to low frequencies, whether it is frequency-dependent or independent. A DIY amplifier with a flat noise floor might offer a different subjective listening experience.
Given the multi-frequency input, the amplifier's total distortion is primarily composed of harmonic and intermodulation distortion. The simulation results show that the highest harmonic distortion reaches about -156 dBV, with the minimum signal-to-harmonic ratio being 152 dB. These results demonstrate excellent noise suppression and distortion control, achieving the expected high-performance standards, at least based on the simulation outcomes.
Figure 7，Multi-Sine Wave Input Circuit Configuration Simulation Results
The following lists the different distortion frequencies and their amplitudes generated from the simulation, for reference by those interested:
- 1.5 kHz, -164.3 dBV
- 4 kHz, -159.5 dBV
- 9.5 kHz, -159.1 dBV
- 10 kHz, -160.7 dBV
- 10.5 kHz, -160.2 dBV
- 14.5 kHz, -157 dBV
- 15.5 kHz, -157.0 dBV
- 19 kHz, -157.1 dBV
For those who might not be familiar with these simulation figures and find them hard to interpret, it's worth noting that if you were to input a true 24-bit audio source into the amplifier, the output would be virtually undistorted. The harmonic distortion is more than 7 dB lower than the 24-bit quantization noise, making it effectively distortion-free.
In addition, I added negative feedback to the circuit from Figure 7 and adjusted the overall gain to 20 dB, 6 dB lower than without feedback. Observing the results in Figure 8, no frequency-dependent harmonics were observed, but the overall noise floor increased by about 1 dB compared to the no-feedback case.
This phenomenon may differ from traditional Class B amplifiers. Class D amplifiers, due to their modulation frequency, generate out-of-band noise. When negative feedback is introduced, this out-of-band noise can also be fed back to the input. While negative feedback effectively suppresses harmonic distortion caused by nonlinearity, the feedback of out-of-band noise may increase frequency-independent distortion, thus raising the overall noise floor.
Although this analysis partially explains the rise in noise floor, it has not been rigorously validated and is provided for reference only. If I gather further experimental data or validation, I will share it separately in the future.
Figure 8， Simulation Results of Multi-Sine Wave Input Circuit Scheme with Global Feedback
Additionally, when observing the operating frequency of the amplifier, it becomes apparent that as the peak value of the input signal increases, the divergence of the main frequency enhances, while the peak itself becomes lower. This phenomenon is a characteristic of the sigma-delta self-oscillation architecture, providing significant advantages in electromagnetic compatibility (EMC). Since the self-oscillation frequency is not fixed, it dynamically changes under different load and signal conditions, dispersing EMI (electromagnetic interference) energy rather than concentrating it at a specific frequency. This reduces interference with other devices, making sigma-delta amplifiers relatively superior in EMC performance in practical applications.
The following text is excerpted from "Audio Power AMP Design Handbook 4th edition" by Douglas Self, which discusses the increasing distortion characteristics of traditional Class B amplifiers at higher frequencies. Although this book was published in 2006, and advancements in technology may have significantly improved the high-frequency distortion characteristics of current amplifiers, this fundamental principle still holds unless there are breakthrough developments.
From the simulation results, Class D amplifiers also adhere to this pattern, but their high-frequency distortion is only a few dB higher than the low-frequency distortion. Given the overall low distortion rates, this difference can generally be ignored.
Another noteworthy aspect is that the amplifier's phase response tends to be linear. As we know, the human ear is not very sensitive to phase response in sound. Therefore, it raises an interesting question: do linear-phase amplifiers produce different subjective listening experiences? If anyone has personal experiences or insights on this, it would be great to share and discuss!
Based on a simple calculation of the simulated distortion effects, the amplifier's THD value can reach approximately 0.0000025% with the minimal SNR is 152dB. Considering that the harmonic components are several dB above the noise floor, the THD+N value should be in the same range. This simulation result seems almost unrealistically good, so I conducted multiple verification simulations and cross-checked with actual test results, ultimately confirming that the simulation results have a certain degree of credibility. The verification approach was as follows:
- Device SPICE Model Verification: Each device's SPICE model was individually verified by building its recommended test circuit using the model and comparing the results with those provided by the manufacturer. The simulation results matched the manufacturer's data closely, confirming the accuracy of the models used.
- Comparison of Simulation and Measured Parameters: Key parameters of the fabricated circuit (such as the amplitude and shape of the integral waveform, the amplifier's oscillation frequency, and gain) were measured and compared with simulation results, with discrepancies generally kept within 10%. This directly demonstrates the reliability of the simulated circuit.
Of course, the performance results from the simulation represent only a theoretical upper limit; the actual amplifier's performance may degrade to varying degrees due to various factors. A good simulation result is merely a starting point, and the process of building the amplifier is essentially a struggle against these degradation factors.
Writing the threads has been particularly challenging for me. I’ve tried to clearly describe the simulation conditions at the time, but the reality is far more complex than what I've expressed here. If there are any unclear or incorrect points, I welcome further discussion. For now, I will share the results of my two major version of DIY AMP and some subjective listening experiences, so stay tuned!
I have two questions about your figure 3, one technical and one semantic question.
The semantic question is whether it is usual to call this a sigma-delta modulator. I have only seen that term used for clocked circuits, while this is a completely continuous-time, self-oscillating circuit. Then again, I am more familiar with ADCs and DACs than with class-D amplifiers.
The technical question is how this fits in your no-feedback concept. You explained that you want to make an amplifier with no feedback at all and that that does not apply to supporting circuits such as power supplies, but in this figure, there is shunt feedback over the entire amplifier (output filter excluded). Will you later replace this with an open-loop equivalent?
The semantic question is whether it is usual to call this a sigma-delta modulator. I have only seen that term used for clocked circuits, while this is a completely continuous-time, self-oscillating circuit. Then again, I am more familiar with ADCs and DACs than with class-D amplifiers.
The technical question is how this fits in your no-feedback concept. You explained that you want to make an amplifier with no feedback at all and that that does not apply to supporting circuits such as power supplies, but in this figure, there is shunt feedback over the entire amplifier (output filter excluded). Will you later replace this with an open-loop equivalent?
I spent a bit of the time researching class D amplifiers. I found a masters dissertation, link provided. This exceptional article covers class D design in a good amount of detail. Here's the link for those who want to learn about class D in audio applications, the Contents' links work, so use that to jump to your topic of interest (@MarcelvdG - go to Chapter 2 on page 60):
https://webthesis.biblio.polito.it/secure/19168/1/tesi.pdf
On a side note, I also found this article: "Simulation of electromagnetic environment of class D amplifier" and asked my daughter to try to download it through her UNI library portal... and she managed to do so... an exciting and somewhat scary read indeed (pdf attached).
https://webthesis.biblio.polito.it/secure/19168/1/tesi.pdf
On a side note, I also found this article: "Simulation of electromagnetic environment of class D amplifier" and asked my daughter to try to download it through her UNI library portal... and she managed to do so... an exciting and somewhat scary read indeed (pdf attached).
Last edited:
Thank you for your questions.
Regarding the first point, I believe it is acceptable to call it a sigma-delta modulator.
As for the second point, the feedback resistor R3 is applied to the output pulse signal only, which is part of the self-oscillating circuit. However, for the audio signal path, the loop remains open.
4. First Version Circuit Build and Distortion Testing
In the previous post, I shared the results of the circuit simulation, which performed remarkably well, even exceeding expectations. The impressive results made me question the accuracy of the simulation for a time. To verify its reliability, I conducted some basic tests and checks, and the findings confirmed that the simulation data is largely trustworthy and effective.
Now, let’s take a look at the photos of the first version of the amplifier. Figure 1 shows the PCB layout design and a 3D rendering. This is a four-layer PCB design with a fully implemented ground plane. Not only does this improve electrical performance, but it also significantly enhances signal integrity and stability, ensuring the circuit operates reliably.
Figure 2 shows a physical photo of the first version of the amplifier, with the entire circuit board mounted on a heatsink, even though additional cooling is not necessary based on actual usage. To make full use of space, the interior of the heatsink has been hollowed out to accommodate a 1700mAh lithium battery, as shown in Figure 3.
Figure 2， Assembled and Tuned Circuit Board
Figure 3，Battery Placement
Figure 4 showcases the circuit board after several months of iterative tuning and modifications. The image clearly reveals numerous signs of adjustments and changes, reflecting the intensive debugging process that has truly tested my patience and attention to detail. My method involved using two identical circuit boards, labeled A and B. I first concentrated on improving and optimizing board A, then compared its performance with board B. If board A demonstrated superior results in both objective test data and subjective listening experiences, I would proceed to enhance board B accordingly. This cycle of comparison and improvement continued until board B’s performance ultimately surpassed that of board A. I repeated this process multiple times, halting only when further modifications yielded no significant gains.
I would like to express my heartfelt gratitude to several friends who patiently participated in the listening tests throughout the debugging process and offered invaluable subjective feedback. Their support has been instrumental in enabling me to integrate objective test data with subjective listening experiences, allowing for continual refinement and adjustment of the circuit design to reach a higher standard.
Figure 4， Modified Circuit Board
Figure 5 shows another circuit board, which has been fine-tuned for driving a pair of small bookshelf speakers. To my surprise, it not only handles these speakers with ease but also delivers exceptionally high sound quality. My initial goal was to test the amplifier’s power headroom and see if it could effectively drive the bookshelf speakers. If it could drive these small speakers well, then I expected it to have more than enough power for various headphones. However, what truly astonished me was how the sound performance of this amplifier exceeded all my expectations.
During my initial listening session, the most immediate impression I had was the incredible clarity of the sound. Particularly with high-frequency sounds, like those produced by a triangle, the sound seemed to extend endlessly, layered and rich, creating a captivating experience.
When I closed my eyes, it felt as if the auditory experience was coming from more than just two speakers; it was as though multiple speakers were playing simultaneously, with an impressively stable soundstage. The vocals were anchored precisely in a fixed position between the two speakers. Regardless of whether it was the dynamic bursts of the instruments or the simplicity of a chord accompaniment, the position and tonal quality of the vocals remained consistently intact, as if there were a dedicated center speaker exclusively handling the vocals.
Figure 5，The Circuit Board Used to Drive Small Bookshelf Speakers
Now, let’s share some results from the Total Harmonic Distortion (THD) tests. I used the SIGLENT SDG2042X as the signal generator and the KEITHLEY 2015 THD MULTIMETER for measuring THD. Unfortunately, the THD of the sine wave directly output by the SDG2042X varies between approximately 0.001% and 0.006%, depending on the output amplitude and frequency, while the maximum measurement accuracy of the 2015 THD MULTIMETER is only 0.001%.
Figures 6, 7, and 8 display the THD test results at frequencies of 100Hz, 1kHz, and 10kHz, respectively, with corresponding THD values of 0.005%, 0.006%, and 0.003%. Due to the limitations of the testing equipment, we cannot precisely measure the actual distortion level of the amplifier. However, based on these results and the characteristics of the measuring equipment, it can be reasonably inferred that the amplifier’s actual distortion level should be lower than the measured values. This indicates that the amplifier performs quite well in terms of distortion across different frequencies, although there is still room for improvement.
Figure 6， THD test of 100Hz
Figure 7， THD test of 1kHz
Figure 8， THD test of 10kHz
I originally planned to share some subjective listening impressions, but given that quite some time has passed, my memories of the specific performance of this version are a bit hazy. Additionally, I didn't try out more headphones at that time to further validate the effects. Therefore, I will have to save the discussion of the listening experience for future posts.
In summary, I am quite satisfied with the performance of the first prototype, as both the background noise and distortion were kept at very low levels. However, there is still significant room for improvement in this version. During the debugging and listening process, I discovered several issues, and as you can see, the circuit board has many test wires from the adjustments. This DIY project was a personal challenge for me, so I decided to create another version and include a housing design (I am personally quite interested in structural design as well). In the upcoming posts, I will share the DIY results of the third version, including both the circuitry and the appearance (the changes between the second and third versions are relatively minor, so I will skip over the second version).
In the previous post, I shared the results of the circuit simulation, which performed remarkably well, even exceeding expectations. The impressive results made me question the accuracy of the simulation for a time. To verify its reliability, I conducted some basic tests and checks, and the findings confirmed that the simulation data is largely trustworthy and effective.
Now, let’s take a look at the photos of the first version of the amplifier. Figure 1 shows the PCB layout design and a 3D rendering. This is a four-layer PCB design with a fully implemented ground plane. Not only does this improve electrical performance, but it also significantly enhances signal integrity and stability, ensuring the circuit operates reliably.
Figure 1， First Version Circuit Board Design Overview
Figure 2 shows a physical photo of the first version of the amplifier, with the entire circuit board mounted on a heatsink, even though additional cooling is not necessary based on actual usage. To make full use of space, the interior of the heatsink has been hollowed out to accommodate a 1700mAh lithium battery, as shown in Figure 3.
Figure 2， Assembled and Tuned Circuit Board
Figure 3，Battery Placement
Figure 4 showcases the circuit board after several months of iterative tuning and modifications. The image clearly reveals numerous signs of adjustments and changes, reflecting the intensive debugging process that has truly tested my patience and attention to detail. My method involved using two identical circuit boards, labeled A and B. I first concentrated on improving and optimizing board A, then compared its performance with board B. If board A demonstrated superior results in both objective test data and subjective listening experiences, I would proceed to enhance board B accordingly. This cycle of comparison and improvement continued until board B’s performance ultimately surpassed that of board A. I repeated this process multiple times, halting only when further modifications yielded no significant gains.
I would like to express my heartfelt gratitude to several friends who patiently participated in the listening tests throughout the debugging process and offered invaluable subjective feedback. Their support has been instrumental in enabling me to integrate objective test data with subjective listening experiences, allowing for continual refinement and adjustment of the circuit design to reach a higher standard.
Figure 4， Modified Circuit Board
Figure 5 shows another circuit board, which has been fine-tuned for driving a pair of small bookshelf speakers. To my surprise, it not only handles these speakers with ease but also delivers exceptionally high sound quality. My initial goal was to test the amplifier’s power headroom and see if it could effectively drive the bookshelf speakers. If it could drive these small speakers well, then I expected it to have more than enough power for various headphones. However, what truly astonished me was how the sound performance of this amplifier exceeded all my expectations.
During my initial listening session, the most immediate impression I had was the incredible clarity of the sound. Particularly with high-frequency sounds, like those produced by a triangle, the sound seemed to extend endlessly, layered and rich, creating a captivating experience.
When I closed my eyes, it felt as if the auditory experience was coming from more than just two speakers; it was as though multiple speakers were playing simultaneously, with an impressively stable soundstage. The vocals were anchored precisely in a fixed position between the two speakers. Regardless of whether it was the dynamic bursts of the instruments or the simplicity of a chord accompaniment, the position and tonal quality of the vocals remained consistently intact, as if there were a dedicated center speaker exclusively handling the vocals.
Figure 5，The Circuit Board Used to Drive Small Bookshelf Speakers
Now, let’s share some results from the Total Harmonic Distortion (THD) tests. I used the SIGLENT SDG2042X as the signal generator and the KEITHLEY 2015 THD MULTIMETER for measuring THD. Unfortunately, the THD of the sine wave directly output by the SDG2042X varies between approximately 0.001% and 0.006%, depending on the output amplitude and frequency, while the maximum measurement accuracy of the 2015 THD MULTIMETER is only 0.001%.
Figures 6, 7, and 8 display the THD test results at frequencies of 100Hz, 1kHz, and 10kHz, respectively, with corresponding THD values of 0.005%, 0.006%, and 0.003%. Due to the limitations of the testing equipment, we cannot precisely measure the actual distortion level of the amplifier. However, based on these results and the characteristics of the measuring equipment, it can be reasonably inferred that the amplifier’s actual distortion level should be lower than the measured values. This indicates that the amplifier performs quite well in terms of distortion across different frequencies, although there is still room for improvement.
Figure 6， THD test of 100Hz
Figure 7， THD test of 1kHz
Figure 8， THD test of 10kHz
I originally planned to share some subjective listening impressions, but given that quite some time has passed, my memories of the specific performance of this version are a bit hazy. Additionally, I didn't try out more headphones at that time to further validate the effects. Therefore, I will have to save the discussion of the listening experience for future posts.
In summary, I am quite satisfied with the performance of the first prototype, as both the background noise and distortion were kept at very low levels. However, there is still significant room for improvement in this version. During the debugging and listening process, I discovered several issues, and as you can see, the circuit board has many test wires from the adjustments. This DIY project was a personal challenge for me, so I decided to create another version and include a housing design (I am personally quite interested in structural design as well). In the upcoming posts, I will share the DIY results of the third version, including both the circuitry and the appearance (the changes between the second and third versions are relatively minor, so I will skip over the second version).