I want to concur with what I think Richard is saying here. There are a number of audio performance metrics which currently routinely well exceed human detection thresholds, such as frequency response, dynamic range, and THD. Once that threshold for a given parameter is exceeded, what is gained by ever further overkill. If, say, -80dB of THD is not detectable via human perception with music, of what listening benefit is targeting performance beyond that?
Richard may also be referring to how some levels of THD, which are humanly detectable in distortion detection threshold tests utilizing sine waves, are completely undetectable with music. I'd add the suspicion that a certain amount and character of THD may even be beneficial to the reproduction of music in the home. However, as I said, that's just my suspicion.
Interesting points there Ken.
What I was hinting at was THD measurements only apply to sinewaves and we don't generally listen to them. In this particular application how well the AD811 will perform with music as a stimulus will depend on its IMD performance. There may be a way to work back from a harmonic level to an IMD figure but we'd need to add some fiddle factor for the fact that in an I/V application there will be dozens of sets of image frequencies, not just the baseband frequencies to work out the IMD from. I'm way too lazy (and not particularly hot on the math involved) to do that.
What I was hinting at was THD measurements only apply to sinewaves and we don't generally listen to them. In this particular application how well the AD811 will perform with music as a stimulus will depend on its IMD performance. There may be a way to work back from a harmonic level to an IMD figure but we'd need to add some fiddle factor for the fact that in an I/V application there will be dozens of sets of image frequencies, not just the baseband frequencies to work out the IMD from. I'm way too lazy (and not particularly hot on the math involved) to do that.
You might even say some THD could lead to a more accurate reproduction, since there must be a lot of cases of studio monitoring electronics with highish THD and without it you would not really be hearing what the producer was hearing.
That is where the discussion regrading electronic distortion and music reproduction becomes highly controversial. I will only share my own occasional personal listening experiences (not via my own system, sadly) where the amplification utilized in a certain replay chain was known to have significantly higher measured distortion than other commonly available amplification, yet which produced sound that fooled my ears in to concluding that I was listening to some form of a live acoustic event. I’m not suggesting any system design conclusions about that, I’m just noting that there seems to be more to human distortion perception, and human music reproduction perception than would seem logical.
Regarding the PSU1
Are the values of the 6800uF filter caps and the output 4700uF caps critical?
If 3300uF is what I have on hand, can I increase the value of R3/R4 to say 5R to get the corner frequency back where it should be.
Regarding the output caps, once again, is 3300uF going acceptable for the ouput caps?
I have some 3300uF UHE series so the esr is still 14mOhm.
Are the values of the 6800uF filter caps and the output 4700uF caps critical?
If 3300uF is what I have on hand, can I increase the value of R3/R4 to say 5R to get the corner frequency back where it should be.
Regarding the output caps, once again, is 3300uF going acceptable for the ouput caps?
I have some 3300uF UHE series so the esr is still 14mOhm.
Me : What is the advantage of using +/-15V than say +/-7V ?
abraxalito : the 3rd harmonic distortion will be at least 10dB lower at 15V than at 7V. 2nd harmonic will reduce too, but less dramatically.
Me : So the difference is -110dB or -120dB. You can hear that difference ?
abraxalito : I don't consider THD to be a relevant metric on music.
So going back to the original question, assuming THD is not relevant as you said :
What is the advantage of using +/-15V than say +/-7V or maybe +/-9V, according to you real-life experience ?
Assuming of course that heat dissipation does not count as an advantage.
Many thanks in advance,
Patrick
abraxalito : the 3rd harmonic distortion will be at least 10dB lower at 15V than at 7V. 2nd harmonic will reduce too, but less dramatically.
Me : So the difference is -110dB or -120dB. You can hear that difference ?
abraxalito : I don't consider THD to be a relevant metric on music.
So going back to the original question, assuming THD is not relevant as you said :
What is the advantage of using +/-15V than say +/-7V or maybe +/-9V, according to you real-life experience ?
Assuming of course that heat dissipation does not count as an advantage.
Many thanks in advance,
Patrick
I have no real life experience of running at higher supply voltages. My AD811 was the SO8 version which is severely limited in heat dissipation so I think I ran it at +/-5V.
That is correct. I'd also be interested in others' listening impressions between higher and lower supply rails.
Here is an alternative approach for PSU. This is basically a "silentswitcher" with 2 identical bipolar power supplies converted from a single 3-15V input. With appropriate setting resistors it can be used e.g for +/-12V & +/-5V or 2x +/-12V or +/-15V & +/-5V. Output current is 150mA on each channel so it can work with e.g. Miro's boards. I've used it with TDA1541. Regulators are TPS7A39 which are not ultra-low noise but still much lower noise than e.g. LM317. If powered from a battery or power bank noise on each channel is about 15uVrms from 10Hz-40kHz. Board size is 58x34mm so it makes it possible to build a very compact AD1862 dac.
Attachments
In about 1993, I attended a seminar from one of the application engineers from Analog Devices. They spent some time going over their flagship audio opamp, AD797, and their flagship video opamp, AD811. The AD811 was designed to drive two 75-ohm video cables at the same time, for a total load of 75 ohms - quite impressive. However, I think it was common practice (and printed somewhere) on the AD811 to limit the PS to +-12 VDC to minimize thermal rise when driving a low impedance load. Of course, you should always watch power dissipation in every circuit. He said the AD797 had its best performance at +-15 VDC compared to lower voltages. That included more factors than THD, like slew rate, etc, I think. Anyway, I tried PS voltage comparisons back then (I do not remember how extensively I experimented with it though) and I concluded that sometimes I could hear a slight improvement in sound with +-15 VDC. However, nowadays, in a miro AD1862 DAC, which is the subject of this thread, trying +- 15VDC on the opamp is not a high priority for me. First, the extra cost and extra effort of a third PS. Next, this application is using an opamp for an I/V stage, which is somewhat unique. I think other factors will likely provide more bang for the buck for me, like trying out different FB resistor and capacitor combinations first. For example, in my miro board setup, I liked the AD797 more than the AD811. However, as the Walt Jung article says, which is a link from post 4897 by Zoran (thank you for that) I did not have the optimum setup for the AD811. Walt Jung's article is very interesting, and I may have to build another miro board with the Jung values in it to try out the AD811, which is in some ways is a different animal than a VFB opamp. So I will ask some questions about what is actually going on and optimum values for a VFB opamp I/V in another post, since this post is getting too long.
I like archeology too, but nearer from you, have you tried opa861 before considering opa 811 designs and readed the thread just in case 
.
Again guys, we have already a pcb for that thanks @Vunce .... have you now to go a little beyond, considered a single ad1862 pcb, with the 861per channel for I/V, for a better RF decoupling and a shorter distance between it and the ad1862 for layout reasons. Then a cmos or fet buffer with two transistors or two opa 1655, one for each channel or simply a single opa1656 buffer for both channels outputs?
Cheap, cheap, good good sound, no sorting out hassle of discrete parts. Only better if no socket adaptaters on the pcb.
Plus it could work as headphone amp as well.
Me I am finsihing my tube version thanks a @gaszto gift and friend advices. But for the monney with selected passive parts if you see my posts above in the thread, I bet half on a hand those silicon oaps will be hard Q/P to beat and democratic price for good music to the people... Hapyness is a warm gun
.Again guys, we have already a pcb for that thanks @Vunce .... have you now to go a little beyond, considered a single ad1862 pcb, with the 861per channel for I/V, for a better RF decoupling and a shorter distance between it and the ad1862 for layout reasons. Then a cmos or fet buffer with two transistors or two opa 1655, one for each channel or simply a single opa1656 buffer for both channels outputs?
Cheap, cheap, good good sound, no sorting out hassle of discrete parts. Only better if no socket adaptaters on the pcb.
Plus it could work as headphone amp as well.
Me I am finsihing my tube version thanks a @gaszto gift and friend advices. But for the monney with selected passive parts if you see my posts above in the thread, I bet half on a hand those silicon oaps will be hard Q/P to beat and democratic price for good music to the people... Hapyness is a warm gun
Last edited:
It's unfortunate that the root problem causing so much technical challenge and effort in active I/V stage design isn't the desired 20Hz-20kHz audio signal. It's the DAC's maximally slewing transition slope between adjacent output samples. That very high slew-rate transition is not inherently part of the desired audio signal. It's only an artifact of a (zero-order-hold) DAC's quantizer switching discretely from one output level, to the next in rapid succession. Perhaps, as a very rough metaphor, think of the audio signal being a ball that's stored and transported within a cardboard box representing the rectangular envelope of a DAC output sample. The box occupies more volume (consumes more bandwidth) than does the ball it contains. The purpose of the box is that it simplifies storage, and transportation of the ball. To obtain end usage of the ball, it must either be accessed through, or somehow extracted from it's box. Which can be done by clumsily accepting both the ball and it's box, or by gracefully discarding the box, and accepting only the ball. Very rough metaphor, I know.
The DAC's output sample level-transition rate is so fast, that it essentially presents an RF signal to the following I/V stage. The high-slewing transition easily drives many VFB op-amps, especially those designed specifically for audio, into slewing overload. So, CFB video-application op-amps became popular for active I/V (although, I'm uncertain about how true that still is), because CFB op-amps typically feature much faster slew-rates, and so, they are less taxed. I have doubts about the benefit of a relatively small amount of harmonic distortion reduction stemming from increasing op-amp supply voltages, versus the level of distortion possibly being produced by op-amp slew-limiting, no matter the supply rail voltages. However, that's just my speculation. I have seen some clever sample transition slew-rate mitigation approaches suggested before. However, the most common solution approach seems to remain that of brute force I/V stage implementation, applying the fastest slew-rate op-amps to the task. Which, consequently, are A.C. spec. far overkill for handling the desired audio signal.
Last edited:
The mitigation solutions to gracefully discarding the ‘box’, range from simple passive band-limiting of the DAC’s output node prior to the active I/V stage. Although, this approach could suffer complications from placing reactance directly on the DAC’s output node. To a moderately complex, exponential-settling sample-and-hold circuit. To a rather complex, yet very clever, solution suggested by Dr. Malcolm Hawksford. Which essentially raises then lowers the DAC chip’s reference voltage along a sine wave shaped envelope synchronized with the output sample rate. The result is sample-to-sample transitions which are smoothly sine shaped, rather than sharply rectangular.
Thanks Ken. Simple can be better for DIY. Please can I ask:
- What general design approach would work for "simple passive band-limiting of the DAC’s output node prior to the active I/V stage" in order to smooth the step transitions and attenuate US hash?
- In terms of ultra simple is there any value in just a single but carefully chosen capacitor on the Iout of the DAC chip to ground before an IV opamp?
- What about just a software solution for the stepped slew stress? If say 8x oversampling is done would this attenuate the slew rate stress by minimising the individual step jumps by 1/8 and elegantly resolve the issue?
My own approach to this is to use, as a minimum, a CLC filter designed to accept as input a current source, and whose output is terminated with the I/V resistor. This site is a good resource for designing such : https://rf-tools.com/lc-filter/
As Ken's already pointed out, the DAC itself might not be well suited to driving such a filter directly. In particular the impedance seen isn't purely resistive across the band as would be the case with just an I/V resistor, there is some 'ringing up' at the highest frequencies. Because of this, and because a low valued I/V resistor isn't the best choice noise-wise, I use a common-gate connected MOSFET between the DAC and the filter. The FET has the effect of greatly increasing the output compliance range of the DAC chip whilst presenting it a more nearly resistive load. The FET normally needs biassing with static CCSs to make its gm more constant vs signal current - CCSs are mandatory if the DAC has a bipolar output.
Scott Wurcer has championed this approach, the AD797 datasheet shows an example using 2nF to 0V.
I think oversampling is part of the solution, given the difficulty of image filtering when the DAC's run NOS. It doesn't decrease the slew rate though, it reduces the step size. With no intervening CLC filter, the opamp's settling time needs to be taken into account when running at higher sample rates - settling time does not decrease linearly with step size - take a look at fig11 of the AD797 DS.