Rob's Amp tests

[peniku8]

Pmillett Sound Card Interface loopback measurements:

Measurement chain is Fireface UC → SCI (Generator input) → Internal loopback → Fireface UC
The SCI converts the balanced signal of the Fireface internally and outputs either balanaced or unbalanced, which can be switched. It doubles the output voltage, so if I input 2V balanced, it outputs 4V balanced or 4V unbalanced. Voltages seen are after conversion (means the Fireface outputs half the voltage displayed).

Measurements are sorted by voltage ascending.

 

 

 

The 1V measurement looks similar to the Fireface loopback, so I guess the pmillett isn't adding too much distortion.
The pmillett started distorting at about 8.6V output (unbalanced). When I switched it to balanced, the distortion increased a bit, but there was more headroom (presumably twice the voltage).
Here is a measurement of the Fireface at 0dbFS, which produced 13,25V at the Pmillett's generator output, which is about 24,6dbU:

[kipman725]

The level seems quite low.  I would expect the noise to be constant with level so SNR should improve if a larger signal is used.  The Fireface UC is spec: THD+N AD: < -98 dB (< 0.0012 %), THD+N DA: -96 dB (0.0015 %) so you're getting pretty close.*

 

*didn't see your last post I think optimal level for the Fireface will be around 4V from your results.

[peniku8]

11 minutes ago, kipman725 said:

The level seems quite low.  I would expect the noise to be constant with level so SNR should improve if a larger signal is used.  The Fireface UC is spec: THD+N AD: < -98 dB (< 0.0012 %), THD+N DA: -96 dB (0.0015 %) so you're getting pretty close.

What do you mean by low level? The last measurement was taken at 0dbFS, which produced 6,6V at the Fireface's output.

Here is a direct measurement of the FF without the pmillett in the loop (max output 6,6V as before):

Best THD is achieved at 0.5V output, best noise floor at maximum.

Here is the THD/Noise vs output plot:

[peniku8]

And the loadbank causes another casualty.. this time it's a Yamaha CR-600. Fuses are done for, but I hope everything else is still good. Measurements of my Hypex FA123 coming soon. Also ordered dedicated measurement equipment, since I already had trouble measuring the true performance of the Hypex amps... Fireface not clean enough. Getting the Cosmos ADC to use with a Topping D10b. Probably also going to get the Cosmos APU once it releases, which includes a 1khz notch and a preamp capable to delivering phantom power with lots of gain.

[peniku8]

Well to be fair, I didn't double check and thought I was testing it into 2x8R, but instead the load bank was set to 4+2 Ohm...

[peniku8]

Which of the two styles would you guys prefer? Currently it shows values for 1khz for full range reference and 30Hz for deep bass reference, but I'm not even sure if that's needed. The data table clutters the graph a lot imo

 

Evaluations will additionally have these:

Which will react like this when they're presented with differing values for bust and peak:

[SME]

Wow, these are nice!

If I may make a suggestion for the comparisons of power at different frequencies: plot these in units of dB instead of watts!  Use 1000 Hz as a reference, so it's always "0 dB" at 1000 Hz and will typically be small or negative at the other frequencies.  (So maybe use "+1 dB" for the top of the Y-axis.)  I think this will convey information that is much more useful to the reader as it clarifies exactly out much output you lose at some lower frequency.

The formula to calculate this is as follows:  dB(f) = 10 * log10( P(f) / P(1000) ) where dB(f) is the decibels at frequency 'f', and P(f) is the power at that frequency.  dB(1000) should always be 0.

Other graphs may also be more useful in decibel units, but decibel units only make sense when you have a meaningful reference point as above.  When comparing different amps to one another, there isn't a good reference unless you choose one of them to compare everything else to.

What do you think?

[peniku8]

13 hours ago, SME said:

Wow, these are nice!

If I may make a suggestion for the comparisons of power at different frequencies: plot these in units of dB instead of watts!  Use 1000 Hz as a reference, so it's always "0 dB" at 1000 Hz and will typically be small or negative at the other frequencies.  (So maybe use "+1 dB" for the top of the Y-axis.)  I think this will convey information that is much more useful to the reader as it clarifies exactly out much output you lose at some lower frequency.

The formula to calculate this is as follows:  dB(f) = 10 * log10( P(f) / P(1000) ) where dB(f) is the decibels at frequency 'f', and P(f) is the power at that frequency.  dB(1000) should always be 0.

Other graphs may also be more useful in decibel units, but decibel units only make sense when you have a meaningful reference point as above.  When comparing different amps to one another, there isn't a good reference unless you choose one of them to compare everything else to.

What do you think?

I record the amp's output as WAV, so I'm logging the db values in the first place, which I then use to derive the powers from. Changing out the values with db values shouldn't be a problem at all, but like you said, if people wanna compare amps at certain frequencies (say you're looking for an amp for your tactile system, so you look at the 8Hz results), you might be better off comparing absolute powers over relative db loss, which you then have to backtrack to compare.

For me it's easier to make out the db changes in my head (like when I'm seeing 1000W at 1khz and then I see 800W at 30Hz that's somewhere around 1db loss), than to reverse the process and I like dealing in absolutes, because we're comparing different amps in the big picture. I could maybe also display both, but I already had enough trouble with Excel to make this work as I wanted it to (it's a universal file and the graph needs to adapt to mono blocks, 4 channel amps etc...), and I think even more info would clutter the graphs a bit too much.

[peniku8]

On 11/28/2021 at 10:09 PM, peniku8 said:

And the loadbank causes another casualty.. this time it's a Yamaha CR-600. Fuses are done for, but I hope everything else is still good. Measurements of my Hypex FA123 coming soon. Also ordered dedicated measurement equipment, since I already had trouble measuring the true performance of the Hypex amps... Fireface not clean enough. Getting the Cosmos ADC to use with a Topping D10b. Probably also going to get the Cosmos APU once it releases, which includes a 1khz notch and a preamp capable to delivering phantom power with lots of gain.

Amp is up and running again, were just the fuses. 45 year old amp takes accidental 2ohm clipping like a champ and only causes 1$ of damage. I love this thing! And it's actually good THD, even by today's standards (0.007% 4V into 8R).

I fixed some other things and mondernized the amp while I was at it, fitting it with a C14 socket (replacing the fixed cable) and clamp style speaker connectors. There are LED-sized light bulbs (!) inside the panel that lights up the tuner's signal strength, of which one still works. I put it in the middle socket and it lights up both meters now, it's great.

[Ricci]

How about removing the 15, 30 and 100Hz results. 

Report something like 8Hz ULF, 60Hz Bass, 1kHz Mid?

Maybe 16Khz Treble?

[peniku8]

On 12/3/2021 at 11:22 PM, Ricci said:

How about removing the 15, 30 and 100Hz results. 

Report something like 8Hz ULF, 60Hz Bass, 1kHz Mid?

Maybe 16Khz Treble?

Yea, seeing how the results on the Hypex amp are barely any different at all, I guess we can safely classify and dumb down application types by a single frequency. 60hz for mid-bass, since that's where the kick drum will sit in most non electronic modern music, 10Hz for the HT ULF crowd and 1khz as general full range guideline, but I'm not sure if 16khz is of any value at all. Music is typically mixed to drop off by 3db per octave and then even more when approaching the last octave, so 16khz demands will already be only a fraction of even just 1khz, unless you're talking long-throw line array applications where you need lots of distance loss compensation, in which case you're most likely using specific system amps anyways (or you somehow need maximum power into a super tweeter!??).

You can call the segments 'Full range performance', 'Bass performance' and 'Deep bass performance' I guess.

[Ricci]

On 12/4/2021 at 8:21 PM, peniku8 said:

Yea, seeing how the results on the Hypex amp are barely any different at all, I guess we can safely classify and dumb down application types by a single frequency. 60hz for mid-bass, since that's where the kick drum will sit in most non electronic modern music, 10Hz for the HT ULF crowd and 1khz as general full range guideline, but I'm not sure if 16khz is of any value at all. Music is typically mixed to drop off by 3db per octave and then even more when approaching the last octave, so 16khz demands will already be only a fraction of even just 1khz, unless you're talking long-throw line array applications where you need lots of distance loss compensation, in which case you're most likely using specific system amps anyways (or you somehow need maximum power into a super tweeter!??).

You can call the segments 'Full range performance', 'Bass performance' and 'Deep bass performance' I guess.

Sounds like a solid plan to me. K.I.S.S.

[SME]

Is it too much to suggest adding back 20 Hz?  I mean 10 Hz is real important to the ULF crowd, but I believe some amps do start to lose a lot of output by 20 Hz, and this would affect a lot more typical sub applications.  So then it would be 10 Hz (ULF), 20 Hz (Deep bass), 60 Hz (Mid Bass), and 1000 Hz (Mid).

I agree that power testing at 16 kHz is probably not important, and in fact, some amps may activate the protection circuit with high amplitude UHF inputs.  More useful for assessing UHF performance would be frequency responses at different impedances, which will often vary for Class D designs.

[peniku8]

3 hours ago, SME said:

Is it too much to suggest adding back 20 Hz?  I mean 10 Hz is real important to the ULF crowd, but I believe some amps do start to lose a lot of output by 20 Hz, and this would affect a lot more typical sub applications.  So then it would be 10 Hz (ULF), 20 Hz (Deep bass), 60 Hz (Mid Bass), and 1000 Hz (Mid).

I agree that power testing at 16 kHz is probably not important, and in fact, some amps may activate the protection circuit with high amplitude UHF inputs.  More useful for assessing UHF performance would be frequency responses at different impedances, which will often vary for Class D designs.

Sure, I can do that. It's still 2 tests less than my previous set up.

My Sanway amp actually mutes when given high amplitude HF information (above 10k), and also muted when trying to run full power into 1R, so at least some protections are working there. Just unfortunate that they don't have thermal protection, which made mine blow up. I mean, it was fan modded and on a 1 minute full output test, which was a bit unfair, but still...

[peniku8]

New ADC arrived today, mind the new scale.

Fireface in comparison (distortion seen comes mostly from its ADC):

From 98db THD+N to 120db THD+N and dropping the noise floor by 20db, that's something I like to see! This is currently limited by the DAC, which will eventually be upgraded too, but I'm pretty happy with this so far. Certainly did not expect having to increase the graph range to -180db because half of the noise floor would drop off the chart even if it was set to -170! At least I can properly measure stuff now without having to worry about optimizing gain structure in my measurement chain (Fireface was distorting more starting from -10dBFS out and/or in). Pmillett measurements following.

[kipman725]

The COSMOS ADC is amazing I have been tempted to get one myself but I recently got a TASCAM US HR 2X2 which I have measured loop back ~103 dB SINAD so I shall see how far this takes me.  The TASCAM MIC EIN is specified as -128dBu or lower so I should be able to measure low noise levels with it as well.

With the fireface loopback measurements you have a THD+N vs level graph, is this from REW? how do you do this?

I like the graphs, preference to the data table I also like SMEs idea of referencing everything to the 1kHz result.

I don't know if your interested in adding more tests but amps like the Powersoft X4 are very thermaly compromised for their peak power output and can only sustain full output in the long term with high crest factor signals (I can't full recall but around 14dB?).  There are tests in production partner showing output collapsing over the course of minutes.  This gets into long term power testing though which you have already shown is risky.

[peniku8]

35 minutes ago, kipman725 said:

With the fireface loopback measurements you have a THD+N vs level graph, is this from REW? how do you do this?

RTA→Stepped Sine→select Step level, like this

And no, I'm not keen on doing more sustained testing. I'm tired of blowing stuff up. It's annoying and costly. If you really wanna test PA amps you could put em into an incubator at say 40°C/105°F to simulate a hot summer day and give it a beating over an hour. Pretty sure most amps will die like that.

[Ricci]

I used to use stepped sine tests for distortion testing of subs. Quickly realized that's a bit too brutal at the maximum drive levels I was using. Cooked a few learning that the hard way. 

[peniku8]

22 hours ago, Ricci said:

I used to use stepped sine tests for distortion testing of subs. Quickly realized that's a bit too brutal at the maximum drive levels I was using. Cooked a few learning that the hard way. 

Yea, even at the shortest setting the signals are still half a second long. That's pretty brutal.

[Ricci]

I was using STEPS at the time and using a really tight frequency interval. I once had an LMS5400 that the motor got so hot you could fry an egg on it afterward. That driver survived probably due to the extra heat sinking of the metallic former and cone, shorting rings etc. Lesser ones did not. 

[kipman725]

Did some quick tests on the TASCAM US-2X2HR similar the the Fireface at 4V loopback THD+N = -98.5dB