I'm amazed that his audio DAC outputs signals up to 2.7MHz to actually hit the 24V/us slew rate of the OPA1656 (1 Vrms output). But if you'd rather buy the OPA2156, more power to you!
The OPA1656 has 6dB more open loop gain at audio frequencies, hence the better distortion. The OPA2156 has amazingly good distortion as well, you just might want to load the output a little less than a 1656. This isn't usually a problem in I/V stages. I always like to maximize the gain in an I/V stage, since is provides a major SNR benefit to the system. Just watch the input impedance of the circuitry that comes next. I'm not sure if a composite is really necessary for I/V.
johnc, you're one of the most astute contributors on the diyaudio forum, so I'd like your opinion---what is the upper ceiling of slew rate, beyond which increases are not audible?
I've been taught that you need 10x the bandwidth to effectively pass a square wave---so a 20K Hz square wave needs 200K Hz bandwidth, a 5µSec period. To slew to the 17.75 voltage peak, the initial rise would need to get there in 1.25µSec, a slew rate of 14.2 volts/µSec. That seems to me to be the most you'd ever need.
I've been taught that you need 10x the bandwidth to effectively pass a square wave---so a 20K Hz square wave needs 200K Hz bandwidth, a 5µSec period. To slew to the 17.75 voltage peak, the initial rise would need to get there in 1.25µSec, a slew rate of 14.2 volts/µSec. That seems to me to be the most you'd ever need.
I would add that a 20 kHz square wave is an "illegal" signal for digital playback, and you're unlikely to find a recording chain that will give you anything close.
I think the point is that asking for 20khz square wave is misleaded request. In 20khz bandwidth there's only 20khz sinewave. I am not sure why he chose the number 20khz for this.
Thats correct , based on my research , most current DAC outputs benefit from a higher slew rate . I am unable to see much difference in our AP but multiple listening tests show a clear benefit .
It reminds me of the superreg listening impression from your website , where there was not a big correlation between test data and what was sounding good.
It shows that sometimes sound preference is more than test data ?
I think for I-V it's more the edge rates that are of concern. Obviously the first opamp after DAC is the critical one. Below I have a pic of PCM1794(2) OP
current 'stepped' waveform. The edge rise time looks around 1nS. I'm not sure what the BW of CRO used was so they could be even faster.
It is a possible reason why there is such a large variance in distortion between various opamp based I-V architectures. For example I have seen
plenty of PCM1794 implementations with distortion not much below -100dB. But I have also seen the same DAC having all distortion spikes below -125dB
with careful attention to first I-V stage RF filtering, the first pole being passive.
TCD
Attachments
Thanks for sharing. What do you think of the requirements for newer DAC like 4499? How much more difficult to get it right?
I'm not sure about AK4499 but my best guess is it would have less out of band
noise than the 1794/2. It has larger current swing requirements so OPA1656
may work really well for this DAC.
TCD
And I would bet that the amplitude of each step is too small to actually reach the slew rate of the amplifier. Remember, to actually hit the specified slew rate of an op amp, you have to create a large enough voltage between the op amp inputs to direct 100% of the tail current into the compensation capacitor. Just because you have a fast edge doesn't matter. You can put a 1mV square wave with an infinitely fast edge into an op amp all day and never exercise its slew rate. It will remain in small signal operation where its rise time probably 0.35 / bandwidth (single pole system).
Here is my frustration with the slew rate conversation around op amps. I think what we really need for audio op amps, is not necessarily high slew rate but rather linear input stages. What I mean by linear input stages is ones that have minimal variation in their transconductance for a wide range of input differential voltages.
It USED to be true that an op amp's slew rate was a fairly good indicator of input stage linearity. Op amp manufacturers would degenerate the input pair and so they could stabilize the op amp with a much smaller compensation capacitor, and the resulting slew rate of the op amp (determined by tail current and comp cap size) was much higher. This is why older high speed op amps and amplifiers targeting video applications have fairly high broadband noise, it is thermal noise from the degeneration resistors.
But all sorts of nonlinear circuits are also now commonly used to increase an op amp's slew rate. These "slew boosting" circuits detect large differential voltages at the op amp inputs, and then dump extra current into the LTP to increase the slew rate during the transient. This is kinda like putting nitrous oxide on a honda civic. You get civic gas mileage (power supply current) during regular usage and all that extra slew rate in a drag race (large square wave with a fast edge). But, depending on how the slew boosting circuit was implemented, this could have drastically reduced the linearity of the input stage. In fact it can even introduce discontinuities into the gm vs Vdif curve.
I'm not anti slew-boost, in fact I have seen it done extremely elegantly to improve the input stage linearity (OPA1692 is an example of this). But I have the luxury of seeing this as an insider, as the customer you have no way of knowing how the datasheet slew rate spec was achieved unless it is so drastic that it shows up in THD curves.
So where does that leave us? For me, beyond a certain threshold, more slew rate is pointless and taken to the extreme it actually makes me skeptical of an amplifier's linearity because I don't know what deal was made with the devil to get all those slew rate bragging rights. It could have a totally non-linear input stage, or an architecture that is going to show low frequency distortion from thermal feedback effects (Class-AB or H-bridge input stages). What is that threshold? A reasonable rule of thumb would be 6 to 10 times the slew rate of the maximum sinusoidal frequency and amplitude expected, e.g. 10 * 2 * pi * f * A.
Or, you know, use JFETs to get a linear input stage and, at the same time, increase the slew rate by a factor of 40X according to James E. Solomon's landmark paper in JSSC. Equation (34) page 322.
(citation)
(citation)
I don't think you can put an exact figure on it. If you calculate the required slew-rate at max voltage and 20kHz, it is smaller than the rule of thumb. But at the limit, although there is no slew rate limiting yet, distortion is already rising. So you want to stay clear of that, add a safety factor. And that is a judgement call.
Jan
Jan
20 K Hz?? That's not what johnc said:
About 14v/µSec at a peak of ~18 volts line level.
Last edited:
Line level vs large signal. Admittedly slew rate is a large signal parameter, and we want to stay pretty far away from being limited by it, but 6-10 or 2-4 or whatever multiple is a "rule of thumb," so don't lose sight of that.
It's very easy to get lost in the weeds. Ask yourself what you're trying to accomplish and the signals you're driving through the circuitry. Are you using your OPA1656 at 18V?! Not +/- 18V rails, signals themselves. Or are you multiplying safety factors on top of pathological edge conditions?
It's very easy to get lost in the weeds. Ask yourself what you're trying to accomplish and the signals you're driving through the circuitry. Are you using your OPA1656 at 18V?! Not +/- 18V rails, signals themselves. Or are you multiplying safety factors on top of pathological edge conditions?