Return-to-zero shift register FIRDAC

To me at least, seems like it should be fine to keep the original voltage reference. Where I suspect there may be a problem is with the IC opamp error amp sound and or X5R cap linearity. IMHO dynamics of music don't sound quite right. Its something not so easy to measure as compared to steady state PSS behavior. Have seen a discrete dac Vref regulator that might help, but it is a proprietary design which I am not at liberty to share. At least with a discrete design it may be possible to fine tune regulator performance in ways that are inaccessible in an IC amplifier.

Also, understood that equally loading both phases of the dac outputs for each channel should help cancel current and voltage variation in the Vref supply. Although I might be able to get some improvement there using approximately equal lumped loading, it might be rather difficult exactly equal the transformer input impedance. In that regard, how close to exactly balanced loading would you consider acceptable?
Last edited:
Thanks Marcel, any idea of the component values to get 5V for Vref and is the output stable when connected to the 22uf cer on the board?
In that particular regulator looks like the reference voltage is developed across a resistor using a current source (at least that's one way of looking at it; actually the current source also needs stable base voltage, but there are various ways that is sometimes done). Or in this particular case, maybe its best to think of the voltage at the base of VT7, VT14 as the reference voltage connection point. Also, its not necessary to use IC regulators as pre-regulators or as current sources (have seen LM317 used either way in this general type of circuit). Circuit can be improved by using discrete current sources instead.
Last edited:
Don't know if I would use the exact circuit of #1,852 if intending to keep Marcel's low noise reference. Also, I wonder if that circuit might have a little too much loop gain for optimal sound dynamics. Don't like having to put a CII ceramic at output. Of course, an amplifier doesn't have to be compensated only at the output. Its just that putting a cap at the output can help with voltage regulator transient response as seen by the load. If instead we use SQ friendly output/bypass caps (linear ones with damping as needed), then the regulator doesn't necessarily have to be compensated entirely at the output. Also IME its possible to over-regulate some audio power rails, depending. Sometimes a little degeneration in a regulator is not necessarily a bad thing for dynamics (also not necessarily a good thing for PSS measurements; IOW a possible tradeoff area in terms of goals).

Sorry if the above seems too vague. Only free to say certain things in some cases, can't get too much into specifics.
Last edited:
Also, I wonder if that circuit might have a little too much loop gain for optimal sound dynamics.
I have to admit it is not clear for me what the above means. As well I have no clear idea what criteria has to match the +5v reg in this topic, sorry. Nazars regs are simple and 2 transistor version is stable and sufficient for many applications.

Don't like having to put a CII ceramic at output.
No problem, put a 100uf elcap there, (just be careful with low ESR in 3 transistor version) and ceramic one at the load then.
Btw, use of LM317 is not mandatory, other regs are fine as well or even unregulated source with Nazars filter.

Another type of semi discrete regs I fancy is Bandgap reference (eg LTC6655LN) - LPF if required- op amp - pass transistor if required.
...what criteria has to match the +5v reg...
Its that the ADJ pin of LM317 runs about 1.25v lower than Vout. The ADJ node is at the same voltage as the base of VT14 (assuming little voltage drop across R36). Also, VT14 emitter must be at 5v, which is what we specify. That means ADJ and VT14 base must be at 4.4v. Since LM317 Vout is 1.25v higher than ADJ, Vout must be 5.65v. To make that happen (and assuming we leave R38 at 240R), then R34 should be set to about 846R. Unless I made an arithmetic error someplace, that should be about it.

...not clear for me what the above means.
Loop gain is how much gain there is around the feedback loop. VT14 is the input of the error amplifier. VT13 is another amplifier stage to provide more gain, and VT12 is the output transistor. However, the output at the emitter of VT12 is connected back to the input of at the emitter of VT14 (like a grounded base amplifier). Connecting the output back to the input (out of phase of course), creates a signal flow loop, a feedback loop to be more specific. The transistors in the loop add gain to input signal as it passes through the loop to the output. That's loop gain. Also, in this case the transistors don't have degeneration resistors in series with the emitters. It means the gain of the transistors is whatever it happens to be for each transistor at its nonminimal operating point, the current of which is set by the 390R resistors. Still, as current goes down a little bit in response to load current draw and voltage variation, the transistors gain goes up or down a little, capacitance can change a little, etc. So long as the variation is small we can use a linear model to approximate things like the gain (a 'small signal' model).
Last edited:
If you read something like Bruno Putzeys' "The F-Word," which is about feedback, as far as the math goes it looks like the more loop gain the better, so long as you can maintain stability (not accidently make an amplifier into an oscillator, that is). Also, spectral analysis tends to look better with more feedback because feedback is used to help linearize the system.

Problems can sometimes occur though. Stability may not be that good under all operating conditions, at turn on, turn off, for particularly for poorly behaved loads, etc. The system may become more sensitive to EMI/RFI when there is huge loop gain, etc. For audio sometimes it just sounds better to limit loop gain and linearize devices locally, say, at each transistor, rather than try to linearize everything at once. According to some mathematical modeling, everything at once should be better, but real world devices may not exactly match a simplified mathematical model. Also, spectral measurements may not capture everything that is going on in a way that gives enough useful insights into system operation, and so on.

One thing typical (PSS) spectral measurements don't do well is measure dynamic system behavior. If you have a fixed level, fixed frequency sine wave then its easy to see if there is harmonic distortion because you can see spectral lines that aren't supposed to be there. However if you modulate the volume level up and down then you are creating a lot of new frequencies by doing that. Its no longer easy to see if there are spectral lines that shouldn't be there. Its starting to get too complicated to easily make useful sense of what you see. Therefore we don't have a great way of measuring what an amplifier does under dynamic (non-PSS) conditions, such as, for example, where there are musical instruments with changing volume levels. We often just assume that if the PSS measurements look okay, then everything else is probably okay too. Not always the case though.

If you want to know about about the sorts of things you can look at to check for trouble in amplifiers, Bob Cordell's book on audio amplifier design, 2nd edition, has a whole chapter on advanced sources of distortion (ch. 16?). I won't say it covers everything that can happen with dacs or similar, but pretty comprehensive for many or most other audio amplifier areas. Between that book and Doug Self books, you can get a pretty good idea of conventional thinking around audio design.
Last edited:
You can do subtractive tests on an amplifier, like QUAD / The Acoustical Manufacturing Company already did decades ago. Those allow you to listen to the distortion on real music with the music strongly attenuated.

Phase differences always cause a residue, even when the phase shift is proportional to frequency and therefore represents a simple constant delay. If it is non-linear effects you are interested in, you can include phase and amplitude correction circuits that make the suppression of the music better.

I did a test like that in 1994 and ended up with about 60 dB of attenuation of the music, with the distortion unattenuated. The residue still sounded like music rather than distortion, as long as the amplifier didn't clip.
For that sine and square test, you better make sure there is no rational relation between the sine wave frequency and the square wave frequency, otherwise it's a periodic steady state measurement. 😲

There is some similarity between the subtractive tests and Bob Cordell's distortion magnifier, but the magnifier was published decades later and was meant for ordinary sine wave distortion measurements. The QUAD subtractive tests involved no active components outside the device under test.