• These commercial threads are for private transactions. diyAudio.com provides these forums for the convenience of our members, but makes no warranty nor assumes any responsibility. We do not vet any members, use of this facility is at your own risk. Customers can post any issues in those threads as long as it is done in a civil manner. All diyAudio rules about conduct apply and will be enforced.

Modulus-86: Composite amplifier achieving <0.0004 % THD+N.

Because in my experience the behaviours of amplifiers when driving difficult loads, in terms of power and current, at high frequencies is very telling. This is the area where an amplifier relying on global feedback has the least available to it, because of the shaping by compensation to maintain stability - from the graphs you've already shown it appears, by extrapolating, that at around 100kHz the performance of the composite will be about the same as a raw LM3886. So, I'm curious as to whether in the real world there is any benefit to your composite approach for IMD behaviour in the audio band when the amplifier is "stress tested" ...

Thanks for responding!
 
My guess is that you will at least get the benefit of the loop gain available at 20 kHz. The unity loop gain occurs at about 1.5 MHz for the Modulus-86, so I'd expect good performance quite a bit past the audio band.

I've been wanting to measure IMD for a while. I may have to wait for work to quiet down a bit before getting to it, however. Stay tuned.

~Tom
 
Tom,

I'm not seeing "0.0004%" on the graphs. Am I not seeing it?
I am seeing things in the range of 0.004 and somewhat below.
Not saying that the performance is bad at all. Just not seeing that.

Afaik, most amps are better in THD at lower levels than high.
And similarly better at lower freqs than high.
Yours seems better at almost max output!

The question of the "first watt" is an interesting and important one.
I'd think that most class A amps and most high-ish bias class AB amps do well there. The good ones, that are already good.

An idea that might be interesting to explore with the 3886 (and given time I would love to try) would be one of the classic (now classic) feedforward distortion reduction schemes. The barrier to these has typically been the cost of a complete amplifier in order to do the trick. Here the cost is nominal.

As far as "better" amps? I think there are a range of designs that reach into this "below 0.001%" THD. Seems to me a few are on here.

Only by way of reference is a random test of my own Symphony No.1 amplifier. Keep in mind that this design is now 25 years old.
 

Attachments

  • BEAR-Labs-SYMPH-No1-Lch.jpg
    BEAR-Labs-SYMPH-No1-Lch.jpg
    358.7 KB · Views: 2,544
I'm not seeing "0.0004%" on the graphs. Am I not seeing it?

Attached is the THD+N vs power plot that I also attached to Post #1. I have annotated it to show <0.0004 % THD+N.

Afaik, most amps are better in THD at lower levels than high.
And similarly better at lower freqs than high.
Yours seems better at almost max output!

You are correct that, in theory, amplifiers distort less at low signal levels than at high. However, in practice, THD+N is measured. Recall, THD+N = Total Harmonic Distortion + Noise. The noise floor is independent of signal level. When the signal level is decreased, the harmonics fall below the noise floor. The result is a rise in THD+N at lower power levels. You'll notice the same on your amp.

Your Symphony-1 is nice, but my Modulus-86 is nearly an order of magnitude better... :)

~Tom
 

Attachments

  • MOD86_0p00035THD.png
    MOD86_0p00035THD.png
    95.2 KB · Views: 2,523
How does one measure the vanishing low distortion of the opamp precceeding the 3886, since the noise floor is not likely to move... I ought to know this.

The THD of the LME49710 is measured by reducing the loop gain of the amp in the test circuit. It is assumed that the distortion tracks inversely with the loop gain. This is described pretty well in the LME49710 data sheet, actually.

What might be confusing things is that the other graphs seem to show 0.0007% @ 1kHz 25W and 0.003% @ 1kHz 1W.

For the THD+N vs POWER, the measurement is done with a 1 kHz test signal and 22 kHz measurement bandwidth. This lowers the noise floor (the noise is integrated over 22 kHz rather than 80 kHz). Hence, the +N component of the THD+N is reduced and the THD+N lower. In my view, this is a more realistic measure of the THD. With a 1 kHz test signal, the measurement still includes 21 harmonics, so it's more than fair there.

For measurement at 20 kHz, I have to open up the measurement bandwidth to 80 kHz to get a reasonable number of harmonics represented. Unfortunately, this also opens up the noise bandwidth. There's no way around that... The nearly 4x larger bandwidth means that the integrated noise is now, roughly, 4x higher. Hence, the +N part of the THD+N is higher and the resulting plot shows higher THD+N. In the THD+N vs FREQUENCY plots, I have to use 80 kHz bandwidth to get an accurate measurement at 20 kHz. That's why the THD+N vs FREQUENCY plots show higher THD+N than the THD+N vs POWER plots. It's the difference in the noise floor that you see.

It would probably help if I showed a spectrum of the harmonics (assuming they even pop up above the noise floor). I'll add that to my to-do list.

~Tom
 
I noticed I forgot to include a title/legend on the oscilloscope shots of the transient response in Post #19. Please find annotated scope shots attached.

~Tom
 

Attachments

  • Transient_1nF_8R_600px.jpg
    Transient_1nF_8R_600px.jpg
    122.2 KB · Views: 2,410
  • Transient_10nF_8R_600px.jpg
    Transient_10nF_8R_600px.jpg
    120.7 KB · Views: 2,334
  • Transient_47nF_8R_600px.jpg
    Transient_47nF_8R_600px.jpg
    121.8 KB · Views: 2,290
  • Transient_100nF_8R_600px.jpg
    Transient_100nF_8R_600px.jpg
    122.8 KB · Views: 721
So, as the power goes up, the dynamic range increases WRT the noise floor. So if the amplifier distortion does not increase in concert with the power level, one can see what is an apparent decrease in THD+N with increasing power.

Have you tried the notch filter technique (null the fundamental) to see if you can dig down lower in the distortion measurements?
 
Last edited:
But there comes a point where one should think, "Is there a point to knowing this?". If the combination of the sensitivity of the typical speaker system and the THD+N of the amplifier at a certain power level means that any artifacts are always well below the residual noise of the listening environment, then any of that distortion will always be inaudible.
 
So, as the power goes up, the dynamic range increases WRT the noise floor.

The dynamic range is constant, actually. The dynamic range is the "distance" (in dB) from the highest signal component reproducible to the noise floor.

So if the amplifier distortion does not increase in concert with the power level, one can see what is an apparent decrease in THD+N with increasing power.

The raw THD should be signal level dependent. However, at low signal levels, the distortion components are well below the noise floor. Hence, THD+N shows "high distortion" when in reality, it reflects the signal-to-noise ratio.

Have you tried the notch filter technique (null the fundamental) to see if you can dig down lower in the distortion measurements?

To my knowledge, the analog analyzer in the Audio Precision SYS-2712, that I'm using, uses the notch method. To measure 0.0003 % THD+N still requires 110 dB dynamic range. There's just no way around that. That's the challenge.

~Tom
 
Its still pretty sweet to see that out of an amp that:
1. Uses lm3886
2. Is, or seems to be, plug n play as far as ease of build
3. Needs no expensive/rare/matched components

You hit the nail on the head. The Modulus-86 easy to build, uses components that are readily available, and provides stellar performance. That's basically the value proposition here. :)

~Tom