Recently I finished a prototype of the FPGA based DAC with following features:
- Singe optical (TOSLINK) input, accepting 44.1, 48, 88.2, 96, 176.4 and 192k sample rate.
- Stereo outputs L/R 1.06Vrms 0dB
- Digital volume control 64 steps of 1.16dB.
- PCBA 160x140mm, 4 layers
- Dual frequency SI552 VCXO clock oscillator 33.868MHz/36.864MHz
- Jitter suppression -100dB at 10Hz, using digital PLL with 0.03Hz bandwidth
- FPGA - XC6SLX9 LQFP144
- Power supply unit - USB-C PD input 15V 5W, isolation 5kV 18pF.
- There is no DAC IC, the audio output implemented as a 16-tap differential FIRDAC + 6-order analog reconstruction Low Pass Filter
The parameters:
ENOB - 17.2 bits
SNR - 104dB
THD - 0.000004%, almost unmeasurable at least -142dB to carrier at 0dB.
SFDR - 140dB minimum.
See attached spectrum of 1KHz sinewave of 1.05Vrms (0dB level).
-30dB notch filter followed by amplification of +36dB was used,
so THD was 32 times better than measured, noise floor and harmonics were -30dB lower.
Attachments
I do not havе an option to disable dithering. It is always 1TPDF for interpolation and 0.25RPDF for the modulator.
The modulator is of 8-order with dual sided PWM output at 2304kHz. The performance is slightly better at 44.1/48k due to better SNR of the Cosmos ADC+APU, used for measurements.
There is no visible difference in the noise floor when the amplitude is changing from 0 to -60 or even to -130dB.
The linearity was the goal, rather than SNR.
The SNR is completely defined by the noise of analog reconstruction LPF. It is possible to improve it to 120-126dB,
but I don’t see any good reason for doing that. Music even at -75dB is not hearable.
The modulator is of 8-order with dual sided PWM output at 2304kHz. The performance is slightly better at 44.1/48k due to better SNR of the Cosmos ADC+APU, used for measurements.
There is no visible difference in the noise floor when the amplitude is changing from 0 to -60 or even to -130dB.
The linearity was the goal, rather than SNR.
The SNR is completely defined by the noise of analog reconstruction LPF. It is possible to improve it to 120-126dB,
but I don’t see any good reason for doing that. Music even at -75dB is not hearable.
So far, I haven’t made a decision yet. I spent a lot of personal time developing this project and I am almost of the age of retirement. It would be nice to have some small income out of it in the future.
Attachments
Good going!! If your basic idea is really sales, I think you shall move the thread to the commercial part of the forum. If not, why not share some of the design documents and or pictures? How much did the board cost for you to build? iIf you where to do a group buy here at the forum, what would the cost be for a member?
//
//
IME you are quite wrong about that. Do you have an references to support that claim?
BTW, I mean this in a friendly, helpful way. But if you have something you might want to make some money from, who do you think might want to use it or be able to use it? If it is for people to listen to, you need to make sure it sounds better than the competition at whatever its cost is. If you want to see it used as a measurement instrument then you need to understand what people are looking for in that market. The way it looks right now is you don't know who might want it or why.
Also, I think you misunderstand what some humans are capable of hearing. Its been known for decades that early CDs without dither would make quantizing distortion audible to some fraction of the population. That distortion was down at somewhere around -93dBFS. OTOH, a brass band buried under other music was reportedly inaudible at -60dB. What's the difference? Let's just put it this way for now: Human hearing is very complicated and there are known thresholds of audibility for some things, such as distortion, for example. The thresholds are not hard limits, rather they are estimates of the middle of a bell curve distribution. IOW they represent and average ear, not one even slightly above average.
Changing the the subject for a moment, its not clear if you know how important clean Vref and how important low close-in phase noise is to critical listeners. If you don't want to cater to that part of the market, fine. But if you want to make money selling upscale product for people to listen to, you need to understand that market. Briefly, people want good sound. Bruno Putzeys made some dacs that sell for about $12k and people say they sound a lot better than a Topping. Where would you want to be along that continuum?
EDIT: As your project seems to be a FIRDAC, may I ask if it is RTZ? Also, how did you chose the resistors for the FIRDAC array? Reason I ask it because I think there may be some people quite a bit ahead of you already here. In relation to that, you may have lowest cost solution so far but it won't be very popular if it doesn't sound good enough in proportion to other options. So, I would ask if you have done or had done for you any listening tests?
Last edited:
ska
Congratulations for your creation
Specs in your 1st post look very-very good.
I think if you increase the output to some 2Vrms and improve the SNR as you say, you will have won the numbers game.
If you add a USB input, you' make your DAC as versatile as most here.
George
I can’t comment on the deeper technical stuff, but this board is not diy-able, but could be sold as a fully built component eg on Etsy like Xrk does. It’s features are not yet sufficient though.
My wishlist:
My wishlist:
- Must be 24-bit input capable
- Add multiple switchable inputs (rca, usb, spiff, i2s) or simply have i2s only and use 3rd party solution
- If not already possible, allow for builders to use their own power supplies, with separate ones for the digital and analogue sections
- Add update port for the fgpa if possible
It's truly fascinating🙂. I'm curious about the reasoning behind the clock of the modulator. Why do you use 36.864MHz (1.5x24.574MHz)/16 = 2.304MHz instead of 24.576MHz?
Additionally, I've never come across the modulation method called "dual-sided PWM." If you don't mind, could you provide a brief overview of its principles?
I don’t see any obstacles, such as BGA chips, etc.
Everything I see can be soldered at home without any problems - You only need a fairly decent soldering iron (T12 with mini-wave or similar) and preferably hot air.
Yes, there are a lot of components, but are there really fewer of them in other DACs?
Alex.
P.S. I’m not sure that the FPGA firmware will be free, perhaps the author will sell boards with populated FPGA and firmware, and user will solder the rest themselves.
Just my own experimentation during development of the Digital Volume Control. It turns out that 0..-75dB was more than enough for home listening.
I have passed my hearing test recently, it is still OK.
My design provides extra 30dB of dynamic range (~104dB real noise floor ).
The noise generated is pure gaussian, it is transparent for human brain to another 20dB,
So the real dynamic range for human hearing should be about 120-125dB.
I consider it reasonable engineering tradeoff.
You are absolutely right. I have no idea who may be interested in this design. It was based on common sense, rather than record braking or numbers racing.
I'm an engineer, not sales guy.
It is a PWM modulator+FIRDAC, RTZ is built-in by principal of operation. Any set of inexpensive 0.1% Thin Film 10K resistors will work fine.
The distortions are already below -140dB, what else do you need? I tested two versions of the FIRDAC 16taps (10K) and just 8taps(5K). The difference in performance happend to be negligible to my big surprise.
Because I have only this samples of SI552 VCXOs. The design was verified up to 50MHz. It seems that distortions highly depend on the PWMs output frequency which is 36M/16 for this modulator.
It was nice to keep them at unnoticeable level. The 36Mhz seems to be very good compromise.
Double sided PWM is usual PWM, using triangular wave carrier to generate it.
It is NOT "ripple compensated single side" that B.Putzeys implemented in Mola-Mola/Tambaque.
Last edited:
It is always 24-bit, I don't care to cleanup lower bits for 16-bit records.
All mentioned inputs/outputs may compromise the performance of the design by creating possible ground loops, ham or distortions.
For example - how you guarantee proper isolation from the computer or proper jitter performance if you allow I2S interface or home made clock generators?
Adding option of RCA outputs, you may completely spoil the performance of your system. The amplifier's input MUST be differential, any profi will tell you that.
Power supply is already perfect for this design. I can't see any mains or other noise even at -160dB level, and it is very cheap - just any USB-C PD charger
with 15V output will be adequate.
Last edited:
Narrow spectral line skirts would preferable. What do they look like now?
Some info on the measurement at: https://www.diyaudio.com/community/threads/phase-noise-in-ds-dacs.387862/post-7063038
Are you aware of excess noise characteristics of different types of thin film resistors? There is some experimental evidence that not all thin film resistors are equally well suited for FIRDAC use.