Beta Test: Multitone Loopback Analyzer software

[MC_RME]

Hey! The trick would be to use a Peltier element to cool the APU with the ADC used as the heat sink. The heat would transfer from the APU to the ADC. (Mechanically that would be a challenge.)

We might be getting a bit obsessive. Here, I thought I was being crazy for designing and having machined a new front panel for the LPF.

The Peltier element is available for the Cosmos ADC chip from Ivan's shop. The change in THD is not from the PCB or the discrete components.

Correction: The heating element...

 

[BKDad]

The Peltier element is available for the Cosmos ADC chip from Ivan's shop. The change in THD is not from the PCB or the discrete components.

Just a detail. I don't think that the thermostat pcb that E1DA sells is a Peltier element. No point in that. It's just a small micro or similar that gets proportionally hot to maintain the chip temperature at some value (50C?) higher than room temperature. Polysilicon resistors have optimum operating temperature ranges.

At least that's what the pcb assembly I bought from Ivan and Sunny is. Are there other versions?

 

[pkane]

Just a detail. I don't think that the thermostat pcb that E1DA sells is a Peltier element. No point in that. It's just a small micro or similar that gets proportionally hot to maintain the chip temperature at some value (50C?) higher than room temperature. Polysilicon resistors have optimum operating temperature ranges.

At least that's what the pcb assembly I bought from Ivan and Sunny is. Are there other versions?

From what I recall it was just a heater element to warm-up the ADC to around 50degC. Ivan said he found that it reduced the third harmonic a bit. When using the APU with a notch, the ADC third harmonic shouldn't matter as it will get filtered out.

 

[BKDad]

Topping E50 balanced > Cosmos LPF > Cosmos APU > Cosmos ADCiso (with a non-Peltier thermostat heater from E1DA).

I think this is probably good enough for testing power amps and preamps. I didn't bother to optimize all the gains and levels for maximum SNR or best distortion.

 

[MC_RME]

That's indeed more than perfect.

 

[abm0]

Thank you. -Reading..

Edit; Stopped reading. Did @pkane later make a comment to your measurements back then?

Never saw any reaction from him, no. I should probably take the case to his Multitone tool thread, but lately I can't seem to get any more clean-ish results with -70 or -80 dB TD+N like I was getting in November. Everything's all -30, -20 dB TD+N now, I think my PC's more than 10yo power supply is getting terminally uncooperative.

 

[Old_School_Brad]

Never saw any reaction from him, no. I should probably take the case to his Multitone tool thread, but lately I can't seem to get any more clean-ish results with -70 or -80 dB TD+N like I was getting in November. Everything's all -30, -20 dB TD+N now, I think my PC's more than 10yo power supply is getting terminally uncooperative.

Alright. As I see it for now, the matter remains somewhat undecided until it's clear whether the method is valid and if pkane's program is truly suitable for assessing the results in this way. I also admit that I didn’t follow the entire discussion in the referenced thread, so my assessment might be based on limited information.

Nonetheless, the discussion itself is quite interesting.

 

[pkane]

Alright. As I see it for now, the matter remains somewhat undecided until it's clear whether the method is valid and if pkane's program is truly suitable for assessing the results in this way. I also admit that I didn’t follow the entire discussion in the referenced thread, so my assessment might be based on limited information.

Nonetheless, the discussion itself is quite interesting.

Not sure about that discussion, don't remember seeing it. But multitone signal can indeed be used to measure frequency response, as can a log or linear chirp or a simple sine sweep. Multitone Analyzer can derive frequency response from a multiple tone test signal. You can easily compare this to a FR using any other method in MTA itself. What's the controversy? The major difference between these methods is the crest factor of the signal being used, and, of course, the somewhat more coarse sampling of the frequency response in a multitone signal compared to, say, a log chirp.

 

[Old_School_Brad]

Not sure about that discussion, don't remember seeing it. But multitone signal can indeed be used to measure frequency response, as can a log or linear chirp or a simple sine sweep. Multitone Analyzer can derive frequency response from a multiple tone test signal. You can easily compare this to a FR using any other method in MTA itself. What's the controversy? The major difference between these methods is the crest factor of the signal being used, and, of course, the somewhat more coarse sampling of the frequency response in a multitone signal compared to, say, a log chirp.

From what I gather in the linked thread, user @abm0 measures significantly larger frequency response deviations using multitone-derived frequency responses compared to traditional sweeps. @abm0 , please correct me if I’m wrong.

I'm not questioning either method; I'm just curious if there's more to learn from this, so I appreciate you chiming in.

 

[pkane]

From what I gather in the linked thread, user @abm0 measures significantly larger frequency response deviations using multitone-derived frequency responses compared to traditional sweeps. @abm0 , please correct me if I’m wrong.

I'm not questioning either method; I'm just curious if there's more to learn from this, so I appreciate you chiming in.

The level of each frequency being measured in a multitone was at -35dB, from looking at that thread. To compare apples to apples, one should do a log chirp or a sine sweep at -35dB level and see how different that frequency response will look. Differences between a 0dBFS signal response and that of -35dB can be revealing of noise or a low-level non-linearity in the device.

 

[abm0]

The level of each frequency being measured in a multitone was at -35dB, from looking at that thread.

That's one of the mysteries to me in that whole procedure, because if you look at the top-right box where I included it, I was always getting peaks within the 0 to -3 dBFS region, trying to maximize SNR while keeping it below clipping. At a loss for why the response graph was always coming out around -35 dB.

 

[pkane]

That's one of the mysteries to me in that whole procedure, because if you look at the top-right box where I included it, I was always getting peaks within the 0 to -3 dBFS region, trying to maximize SNR while keeping it below clipping. At a loss for why the response graph was always coming out around -35 dB.

Total signal is the sum of all the multitones. Each individual tone in a multitone has a much lower level than the total signal, and that is what's measured for FR chart in MTA.

 

[abm0]

Total signal is the sum of all the multitones. Each individual tone in a multitone has a much lower level than the total signal, and that is what's measured for FR chart in MTA.

Still not sure I understand. If I turned it up by 3 dB it would clip every time, that realtime oscillating "VU meter" bar would turn completely red and I think some text somewhere would tell me it's clipping. Whether you're integrating 300 tones or measuring just one, clipping is clipping, it's still when you hit 0 dBFS anywhere on the spectrum, right? So I first checked where that happens and took down the DAC/amp's volume a few notches from there to do another 'useful run'. What else could I do?

 

[Blumlein 88]

Here is an up close view of two tones. 4000 hz and 4400 hz. If viewed alone you see a simple sine wave. When combined they have a varying level with time interference. At times they are in phase and combine to twice the level and at times are out of phase so they are at zero level even though both sine tones are always there. With two or more tones this might cause your VU meter to oscillate a bit and yes if the level is too high clip. You have to reduce the level of each tone as you add more and more tones. Yet they will at times add up to be nearly clipping.

 

[abm0]

Yeah, makes sense. Still, I have no control over that, the 300 and 400 tone presets are what they are in the tool, all I can do is prevent overall clipping. So it looks like the FR result will always be at or below -35 dB as long as I'm avoiding clipping.

 

[pma]

Correct. That is the reason why multitone does not check for non-linearity related to higher dv/dt, the high frequency non-linearity. Single sine of 20kHz, twin-tone 13+14kHz or 19+20kHz, or DIM30 (square 3.15k + sine 15k 4:1 ampl.) do the job.

Here is an up close view of two tones. 4000 hz and 4400 hz. If viewed alone you see a simple sine wave. When combined they have a varying level with time interference. At times they are in phase and combine to twice the level and at times are out of phase so they are at zero level even though both sine tones are always there. With two or more tones this might cause your VU meter to oscillate a bit and yes if the level is too high clip. You have to reduce the level of each tone as you add more and more tones. Yet they will at times add up to be nearly clipping.

View attachment 438959

 

[DonH56]

Assuming all tones are at the same level A, for N tones each individual tone needs to be reduced by A/N so that when they all sum in phase the signal is not clipped. That means 30 tones would be -29.5 dB each compared to a single 0 dB tone, and 40 tones would be -32 dB each. Allowing a small 3 dB "pad" for intersample overs or just a little margin, you are at -33 to -35 dB for the individual multitone signals. There are games you can play but in general you have to reduce the individual tones' amplitudes to prevent clipping when they all "line up" in phase so all their amplitudes adds.

 

[Old_School_Brad]

Thank you, @pkane @Blumlein 88 and @pma
It's actually quite intuitive, and I associate some of the wave behavior with PID controller oscillations, which I'm familiar with. I’m not entirely sure if the comparison is fully accurate, but it helped me understand your point.

That also means @abm0 should ensure he's not clipping and redo some of his measurements. Since he needs to attenuate the tones, they might end up too low and enter a non-linear range, which would end up affecting the resulting response, is this right?

 

[DonH56]

Low tone amplitudes would in general be more, not less linear, but closer to the noise floor so may require more averaging and/or additional points to reduce the noise in each FFT bin.

 

[j_j]

Hidizs S8 multitone was one example I found a while back. But there's plenty. Almost every time Amir measures an affordable DAC/amp with multitone, the test tones come out with a visibly non-flat envelope, even at the bad zoom level he shows them (because he's only using them as noise tests), with spike heights that cross into potential audibility by having more than 0.1-0.2 dB of difference, yet this goes completely ignored in the results discussion.

In my limited explorations of this, the amplitude differences get even worse when you test with hundreds of tones and large FFT sizes. This requires some investigation, I only got to trying such measurements after hearing DAC/amp differences in (admittedly sighted) A/B comparisons. There could be something there, but I'm not seeing a lot of people discussing it to clarify, only to handwave it away.

There is a whole lot of time/frequency analysis involved here. First, not sure what you mean about a 'non-flat envelope'. Unless the test was carefully arranged to be exactly synchronous, there are a bazillion ways this could go wrong, of course, first I need to know what "envelope" you're talking about. Don't forget you'll see "stuff" if you window off a tone (part of a multitone) in the middle of a peak, due to the FFT window length. Sometimes synchronous (exact same sample rate in and out) is the only way to be sure what you're seeing there.

Offset multitones are very good at finding IMD, as well. I'd be using one for that, and using an all-pass measurement of impulse to get frequency response. All of this is trivial with a good DAC, a good ADC, and a copy of base Matlab plus Signal Processing package. (Yeah, that costs a (*(&*(ing mint.)