Hello, regarding I-V converter opamp at the output of a DAC, it seems that this should have very high slew rate (hundreds of V/uSec) in order to be able to process without distortion the spectral energy in output of the current source DAC (several MHz).
On the other hand, it should be always better to include a big capacitor (2nF or a better CLC filter) between inverting input and ground in order to shunt high frequency energy to ground and limit the opamp to slew too much and, therefore, avoid TIM issues (check datasheet AD797 page 17)
I cannot conciliate the two. Why own a "Lamborghini" if I need to conduct it with the handbrake always on?
On the other hand, it should be always better to include a big capacitor (2nF or a better CLC filter) between inverting input and ground in order to shunt high frequency energy to ground and limit the opamp to slew too much and, therefore, avoid TIM issues (check datasheet AD797 page 17)
I cannot conciliate the two. Why own a "Lamborghini" if I need to conduct it with the handbrake always on?
Also, many DACs have internal reconstitution filters so the output is not as 'hot' as you might think based on the converter frequency.
Jan
The high GBW is nice from a noise gain perspective, as you can shrink the compensation cap in the second leg of the feedback loop. A quick read of Scott Wurcer's notes in the AD797 DS shows a circuit example of a first order passive filter sitting in front of the I/V as well, which can be placed well away from 20k but under any point where we'd start worrying about ultrasonic hash.
Something like the ada4898 is so great for the job, fast and stupid low distortion. Actually any of the modern high end bipolar opamps are fantastic (modern including the aforementioned 797) but one can make an excellent I/V out of a 5532 (and even a little better out of the 5534 if you play with the compensation). Benchmark did for their first two generations before moving to the 49860 (I believe?).
Something like the ada4898 is so great for the job, fast and stupid low distortion. Actually any of the modern high end bipolar opamps are fantastic (modern including the aforementioned 797) but one can make an excellent I/V out of a 5532 (and even a little better out of the 5534 if you play with the compensation). Benchmark did for their first two generations before moving to the 49860 (I believe?).
Unfiltered DAC output is an analog signal made of steps and plateaus. Finite slew rate converts steps to ramps. Is that so bad?
Make sure you check that the amplifier is stable when you do this. Normally putting 2nF of capacitance on the inverting inputs is not a great idea, unless you've also included some capacitance in the feedback path. Especially on the types of op amps that have 100s of volts per microsecond of slew rate.
I'm very skeptical that the amplitude of high frequency content at the output of an audio DAC is anywhere large enough to exercise the slew rate of an op amp.
I've played for a couple of weeks with a Sony CDP 750 which has no low pass filter at all and nothing seemed to be wrong to my ears with an usual lm833 or lm4562 as i/v...I just added a passive low pass filter to be sure there's nothing there and that was all about that. I will try ada4898-2 and 4898-1, then lm6172 very soon and report if there's any positive change.I have a serious feeling that most problems are simply not there...and people are just making unnecessary mods to have a reason to add even distortions to their systems as Pass once mentioned.I think actually that people aren't comfortable with the lack of noises and distortions.So , instead of making unnecessary mods to players, i built my "added-distortion-noise" box and voila, any cd player sounds wonderful now!
And if not stability, throws a huge amount of noise gain at/near the upper edge of audio bandwidth. It's a definite case of mind your p's and q's!
Then cut it at 16khz...no audiophile guy will hear it anyway.I don't even understand why the audio band is defined 20 hz-20khz if almost no professional can hear anything more than 14khz ...I've heard smth about children being able to hear up to 18khz, but they don't even care about the audio components quality.By the time they become passionate and educated about music they are barely able to hear up to 12khz...I will work on a specific system that cuts anything at 15-16khz and see if i can hear anything wrong.Just listening to youtube audio tests i've been able to hear only up to 13.8khz, nothing more.I'd say that my system should be cut at 14khz, but with a little bit of slope , 16-18khz would be the margin. Did anybody tried to make a system that can't reproduce a damn thing over 15khz but good enough at 14khz?
Last edited:
Yup! Mind your Ps and Qs - Wikipedia
In my case I meant it as "pay attention to what you're doing"
My apologies, I sometimes fall into colloquialisms. Across the board, I'm beyond impressed by the mastery of English on this forum.
Yes, because it's a non-linear phenomenon.
That is, assume you have a good multi-bit DAC with a current output. The designers have then done their very best to ensure that the output current steps are proportional to the steps in the digital input code.
Put this current into a slewing transimpedance amplifier and the proportionality is lost, because the steepness of the slewing part is more or less constant, so not proportional to the size of the step you make in the digital input code.
For a DAC without noise shaping, this causes distortion. With noise shaping, it can also mess up the noise floor because out-of-band quantization noise can intermodulate with itself and cause intermodulation products in the audio band.
Mind you, op-amps slew when their input stages are driven into hard clipping, but the input stages will normally already distort substantially before reaching hard clipping. That is, you have to stay far below the actual slew rate limit if you don't want any slewing induced distortion / transient intermodulation distortion.
Last edited:
It depends on the DAC - some DACs are voltage out and already have filtering internally so don't tax the downstream opamps much. Others with current output slew much faster than a typical audio opamp (they are logic devices by and large). Take the DAC I use, TDA1387 (long obsolete but easily available) - it specs a current output settling time of 200nS (to 1LSB) but the edges themselves have a much faster slew rate than that. Normally a VFB opamp configured as I/V is equipped with a cap in the feedback loop so slew limiting isn't encountered.
My listening ability is 15kHz at max. It's reasonable to have a brick-wall filter at 21kHz like red book CD. That's the way I used to. But incredibly, I must admit I can hear the difference between without 21kHz and with more than 21kHz. My system is a digital crossover 4-way. The tweeter is more than 7.2kHz. Someday I happened to design the coefficients without 21kHz cutoff, the 1st pic(the input is real music). The 2nd is with 21kHz brick-wall.
Both had a clear audible difference. I could hear better bass in the 1st pic(more than 21kHz). Some people say if you want bass, improve treble. I'm for this since my several experiences say yes. I'm sure I can clear the ABX test because several instruments have an audible difference in bass. So, my music files now don't have a 21kHz cutoff. I need to remaster my 44.1kHz files to 48kHz. It's not difficult to convert 44.1kHz to 48kHz since the original ones are digitalized from vinyl at 96kHz.
My hypothesis about the "hyper ability" is bandwidth. Removed power by brick-wall isn't small because it's not single tone but bandwidth. I calculate the power more than 21kHz. The 3rd pic shows -81.51dB. So, sound pressure is almost 29dB, which can have a different effect from a single tone. Audible ability on a single tone(15kHz in my case) isn't an optimum probe to decide max frequency. Sound pressure by the bandwidth may be more important for the human ear.
Attachments
Also PCM56 datasheet reports 350ns current output settling time, measured with an active clamp to provide a low impedance for approximately 200ns.
200ns, means 5 Megahertz. Ideally, the circuit's slew rate should be at least 5 (I read even 10) times the maximum output signal's slope for minimum distortion.
Required slew rate= 2VRMS x 1.414 x 2Pi x 5Mhz X 5 times = 450V/us (= AD828 slew rate)
Then we need anyway a small passive network to shunt RF energy to ground in order the ompamp never reaches the above limit (also to preserve stability, but this is another story).
An LM6172 with 3000V/us and being unity gain stable would be a much safer choice as will never have TIM issues in this DAC application.
Is this directionally correct?
The step from 200nS to 5MHz looks to me not to be sound. The settling time in total is 200nS but the rate of change of current at the start of the transient is what will matter if we want to avoid slew-rate limiting.
If you want to do some math, this paper is a good introduction : (PDF) CURRENT-STEERING TRANSIMPEDANCE AMPLIFIERS FOR HIGH-RESOLUTION DIGITAL-TO-ANALOGUE CONVERTERS
Last edited:
Slew rate is only one figure, you need high GBW as well and probably more importantly with the cap in the loop. Jcx, who posts here, has several good posts on this topic.
I'd just use something like ADA4898 and forget about it, but this is one of those topics that will never die. You can use some crazy fast op-amp and sacrifice audio band specs, for sure. If you don't need high output / rail voltages there are a bunch of new fast op-amps that have 5 or 10V supplies.
I'd just use something like ADA4898 and forget about it, but this is one of those topics that will never die. You can use some crazy fast op-amp and sacrifice audio band specs, for sure. If you don't need high output / rail voltages there are a bunch of new fast op-amps that have 5 or 10V supplies.
Last edited:
Assuming a traditional op-amp with high open-loop input impedance on both inputs (that is, not a so-called CFB op-amp):
Very often the feedback network around the op-amp consists of a resistor and a capacitor in parallel, so the transimpedance amplifier (or current to voltage converter, whatever you like to call it) makes the first pole of the analogue reconstruction filter.
With such a circuit, after each step of the DAC output current, the error voltage across the op-amp's input jumps from a near-zero value to the magnitude of the current step multiplied by the open-loop output impedance of the op-amp. Then the feedback kicks in and reduces the error voltage to a near-zero value again.
In this case, you have to ensure that the maximum DAC current step multiplied by the op-amp's open-loop output impedance is smaller than the maximum voltage the op-amp's input stage can handle without significant distortion. Any op-amp of which the input stage is a bipolar differential pair without emitter degeneration can only handle a couple of millivolts more or less linearly, say 10 mV or so. It can be much larger for FET op-amps, especially those that have a high ratio of slew rate to gain-bandwidth product.
Very often the feedback network around the op-amp consists of a resistor and a capacitor in parallel, so the transimpedance amplifier (or current to voltage converter, whatever you like to call it) makes the first pole of the analogue reconstruction filter.
With such a circuit, after each step of the DAC output current, the error voltage across the op-amp's input jumps from a near-zero value to the magnitude of the current step multiplied by the open-loop output impedance of the op-amp. Then the feedback kicks in and reduces the error voltage to a near-zero value again.
In this case, you have to ensure that the maximum DAC current step multiplied by the op-amp's open-loop output impedance is smaller than the maximum voltage the op-amp's input stage can handle without significant distortion. Any op-amp of which the input stage is a bipolar differential pair without emitter degeneration can only handle a couple of millivolts more or less linearly, say 10 mV or so. It can be much larger for FET op-amps, especially those that have a high ratio of slew rate to gain-bandwidth product.