A listening test of two power amps

@erik_squires "1. Match level by multimeter instead of SPL. Get a 60 Hz signal and check the output on either speaker. You can get really accurate this way."

I am a little ambivalent about using 60Hz sine and voltage to verify output level balance. It can tell you something about what an amplifier is doing, and something about speaker efficiency, but I am not sure it can translate into anything about sound quality. There is also the problem of ensuring the meter you are using is measuring true RMS versus average AC voltage. If you are trying to use that method to balance output level, I think you would be better to use pink noise which has a full frequency range. Measurement becomes more of a problem, but it more closely resembles what you may hear when converted to sound using real music. In other words, since speakers are a reactive load, frequency and duration affect response and the amplifier’s drive of voltage across a frequency and time response will vary and may affect actual perceived sound volume.

I don’t really understand why SPL would be frowned upon. It is specifically an agreed measurement of perceptual sound volume reference. It is tailored for a specific frequency range and can be A or C balanced for perceptual preference favoring broad frequency response (A) or lower frequency weighted (C). The only caution for use in this testing is that the ambient noise level is maintained the same, and the SPL is taken from the same location in the same environment. If you are within 0.1 or 0.2 db that is considered well matched.

The anti-testing and anti-measurement crowd certainly makes an impression. It's a good thing those folks aren't doctors or pharmacologists, or we wouldn't have gotten any further than "which medicine do I think makes me feel better..." [which, back in the day, is how people chose which, literally, snake oil product they'd consume]

I think it's a great post and even if not scientifically perfect, this is the type of thing that can actually validate all these claims of magnificent differences in amps, cables, whatever. So I have to assume they don't want to know what the emperor is wearing, or not. 

OP:

It’s your experiment, but as someone who worked quite a bit with microphones and speakers and electronics, the advice I gave you is solid.

Measuring the SPL of quasi random white or pink noise is hard to get precisely accurate, it’s even hard with a multi-meter, which is why level matching should be done with a multimeter and single signal.  60 Hz is a frequency which is convenient as any $10 multimeter will read it, and all are more relatively accurate IMHO than an audio meter, given they are not subject to random room noise or the periodic variability of white/pink noise sources. Level matched experiments comparing DAC’s for instance should be level matched this way and with a decent meter will give an accuracy well better than 0.1 dB.

Now, once we agree to use a multi-meter, and a standard sine wave, whether they measure RMS voltage accurately, peak or peak to peak is irrelevant so long as they are absolutely RELATIVELY accurate. That is, you can trust that 3V now will be 3V in an hour after you switch amps.

Also, with a 60 Hz sine wave signal, again, any $10 meter will measure the voltage pretty accurately and almost perfectly relatively accurately. A nice meter will give you more zeroes.

@surdo 

Finally got to due a little more compare of the e & t files. My first try was at a lower volume than I think should be used for comparison. First pass at that low volume, file e was much better than t, but on replay, t was closer to e than first pass of either. Maybe my electronics were not fully warmed up.

Today played at more reasonable listening level, and also played through headphones. Both files very close overall. Occasional portions sound a little better on one or the other. In general, I like the piano sound (tibre, resonance, attack) better on e than on t, but that is mostly from middle C and above. Lower octaves may be slightly better on t. Bass very close on both. Drums and Drum kit too low in volume and too far in behind piano and bass to judge.

In general, I think most of the differences may be more environmental and microphone, rather than effects of either amplifier and connection. Note especially squeaky wheel/bird song audible on file e from about 4’43’’ to 4’51”, but missing on file t.

So in all, I think you tried an interesting experiment, but I can only tell you about what I can hear differently on the 2 files, and I think there may be more environmental artifacts in the two files than you were hoping for. Some may even be from the microphone and tascam reaching a thermal equilibrium, as well as the electronics and the room. Not sure I would hear the same things if I were in the room listening without the microphone and Tascam involved, so I think any of my comments about the files should be discarded.

@erik_squires "It’s your experiment, but as someone who worked quite a bit with microphones and speakers and electronics, the advice I gave you is solid."

As I said, I am ambivalent regarding matching amplifier out at speaker via 60HZ for amplifier compare. 60Hz sine voltage matching thru the rest of the audio chain is a really good way to compare different pieces of the chain. It is also a good reference for speaker efficiency. For all that is going on between and amplifier and speakers I am not sure it takes everything into account. 

I would love to see a graphing/charting of 60Hz voltage matched amps to the same speakers done in an anechoic chamber with dbA and dbC with some wide frequency music to see the active perceptual effects. Know anyone that can perform this kind of testing? Will see if I can find anything regarding similar testing.

Thank you very much budjoe for taking the time to listen to the recordings and for the detailed account. Yes, it’s a very good point about environmental sound affecting the recording and the listener’s ability to distinguish between the two. There was perhaps a bit more wind in the first recording too along with those angry birds at the end. I think the Tascam and the valve mic had been on for over half an hour and the second test was done shortly after the first. So probably not much of an imbalance there. But good point, also.

jji666. I agree!

Thanks too, erik_squires. I think it’s a great idea to do this test bi-passing the speakers and consequently any unwanted ambient noise and irregular room interactions. I’m hoping I can do this later today or on Wednesday. I’ll likely do the recording at the amp end of the speaker wire. Some recent surgical work has left me a bit shy of shifting equipment!

Back with a new test as per erik_squires’s suggestion

This time the tannoys high and low frequency terminals were jumper-ed and there was no bi-amping or bi-wiring done.

Three recordings were made of three pieces of music with two amps. One amp has two speaker out options, so that’s why there’s three recordings. The three amp configurations are name "j", "y" and "p". The three recordings are of a sparse piano track with a change in dynamic in the middle: "piano". A samba jazz trio: "trio". And a voice with orchestra track (great production but the master recording could be better, i think): "voice". So the 9 edited recordings are:

https://escuta.org/webtmp/j-trio.flac

https://escuta.org/webtmp/y-trio.flac

https://escuta.org/webtmp/p-trio.flac

https://escuta.org/webtmp/j-piano.flac

https://escuta.org/webtmp/y-piano.flac

https://escuta.org/webtmp/p-piano.flac

https://escuta.org/webtmp/j-voice.flac

https://escuta.org/webtmp/y-voice.flac

https://escuta.org/webtmp/p-voice.flac

Steps.

1. A 60Hz sine tone was recorded at approximately -1.0dB

2. The tone was played with each amp setup with the voltage of the amp adjusted with the preamp to measure 0.5V. Fine adjustments of the voltage were made with the volume control of the digital source (Logitech Media Server)

3. For each amp, the three musics were played at the adjusted amplitude.

4. Recordings were made at 48kHz with a Behringer UCA222 interface on a laptop running the DAW Ardour.

5. Nine edited ecordings exported as 48kHz 24bit Flac files.

If anyone has time to listen and give some feedback, please do!

Listening to these now, I think I’ve found an unpleasant problem and not just a difference, in one of the amps....

The problem is in the recordings j-piano.flac and y-piano.flac which are recordings of the Sunfire’s "Voltage source" and "Current source" speaker outs respectively. The piano, in its quieter passages, produces a distinct distortion, especially on certain notes. The recording p-piano.flac, that of the Quad 303, produces no such distortion.

I have a theory on what the problem is:

The Sunfire’s speaker outputs produce a low level, but audible on headphones, 60Hz hum (or a harmonic of 60Hz). The Quad, while perhaps more noisy (with hiss) has no 60Hz tone. I tested both the Sunfire’s unbalanced inputs with the Quad 34 preamp and the balanced inputs with the sound coming from a mixing desk. Both produce the hum and the amp produces a hum even with no inputs connected and with the laptop, that’s monitoring the terminals, running on batteries. I’ve tried disconnecting the Sunfire’s earth and reversing the pins, but the hum is unchanged.

My theory is that when the piano is played softly, certain notes seem to interact with the 60Hz hum, producing what seems like distortion, so if the hum can be removed, perhaps that distortion will disappear.

Is some kind of a mains filter likely to solve the problem or is there something that can be done to the Sunfire to make it less susceptible to this 60Hz hum?

And yes, I do hear the distortion through the speakers. I had noticed this sound before on occasions, but since I was using the Sunfire in my studio, I always suspected the speakers (a pair of JBL 4312A monitors), which have had a long hard life.

Listened to all nine as originally provided, and with a modification (more on that later).

In general across the 3 different versions of each I liked j the most, y second, and p the least. Differences are slight. I like the attack on j and y the best, and think that p tends to round the sound off more; both on the attack and the decay. Most differences showed up in the trio recordings where the lead guitar has a much more natural sound to me. The plucking of the strings and the resonance of them is better and more natural. Also most noticeable in the voice where her voice cracks in the beginning at ”and you know darn well”.

Had a hard time telling much difference in the piano samples except that once again I felt the p was more rounded. That is the only way I know how to describe it. Bass was also very hard to tell across all samples. You may need someone with better sense of piano and bass to hear the differences in those samples.

Here is the more later.

I loaded all the samples in an app that lets me analyze the waveform, and also see there that the j and y samples are very close to each other, and the p ends up being pretty close to 1 dB lower in both average RMS, max RMS, and peak amplitude. across all samples. Within that app I normalized all the samples. Used -3.0 dB peak for the voice and piano and -0.3 dB peak for the trio as that was the max peak in the originals.

I played both the normalized and originals thru 2 different playback apps and my observations hold across both the original and normalized thru both apps.

If you want the full analysis values from the originals and the normalized files, I will send them in a direct message. Let me know.

Just read your post about distortion on the j and y piano files. I think you may have something else going on. Looking at the wave form of the j and p files, there is almost no ripple in the quiet parts on either file. Certainly not enough to be audible. (Wish I could record something from lp to digital with that low a signal, but maybe your source was not lp). I pushed those files (j and p) to the -0.3db normalization and again see almost no ripple in the quiet parts and very little difference between the files.

Were listening to these files through the speakers or headphones? Possible you were hearing issues with the headphones and possibly being overdriven?

"I played both the normalized and originals thru 2 different playback apps and my observations hold across both the original and normalized thru both apps. If you want the full analysis values from the originals and the normalized files, I will send them in a direct message. Let me know."

Thanks a lot budjoe. I really appreciate your comments. Yes, please send the analysis. I’m interested to see exactly what you’re looking at.

"Were listening to these files through the speakers or headphones? Possible you were hearing issues with the headphones and possibly being overdriven?"

I don’t think the phones were overdriven. I hear this sound through the speakers and on headphones (AKG K501 and K246). What I don’t hear on the speakers is the 60Hz buzz. This I only hear when "tapping" the speaker output of the amp and listening on headphones, either directly or on a recording. Although that doesn’t necessarily mean that the sound is not there - there’s a bit of other noise here always.

The more I listen to the sound, the more it sounds like a type of ring-modulation of the piano and I wonder if some sort of cross-modulation is occurring. If this is actually the case, there would be additional side bands appearing in the analysis and not necessarily a distortion.

Listening now to the 3 piano files on phones and at the same volume, p-piano.flac does not have this ring-modulation sound in the first half of the recording, but y and j certainly do.

The original recording is of tracks 8 and 9 ("Var. V. Lento" and "Var. VI Poco movendo") of Alessandro Simonetto’s album "Erik Satie: Works for Piano". I streamed them into a DAW from Qobuz and used this recording for the three amp tests.

budjoe, I normalised the peak of the j-piano and p-piano recordings to -0.3dB (which is what you did?). The above is Ardour’s spectrum analysis of the 1st note of the two recordings. The j-piano recording seems to have a lot more noise. The p-piano part seems to have stronger high frequency transients between 1 and 5kHz.

Also, just above the 1st orange band above 100Hz, there’s what seems like a pulsing partial that’s not present in the p recording. This was evident before normalising both recordings and in which there was less apparent noise but still very weak upper partials (between 1 and 5kHz) in the j recording.

In the samples I sent to you, the waveforms were normalized to -3.0 dB, then the zoomed portion was normalized to -0.3 dB.

In the zoomed section, the noise on the quiet part of the j-piano-zoom was very similar on both the left and right channel. Some high frequency noise, but very low level. In the p-piano-zoom, the left channel (top) has almost no high frequency noise, but the right channel has the most noise of all the waveforms. That is what leads me to believe there is something else going on.

Connections, grounding, or impedance, or some problem in one of the amps. I can’t say from here. I think you should explore some tests with the minimum number of components in each chain. Try with all components in the same power circuit. If any component used has only a 2 conductor power cord, try turning that cable over to see if it changes the waveform and/or the sound. Swap left and right channel interconnects one at a time and see if the noise moves from one channel to the other. Swap the speaker wires from left to right again to see if the noise moves from one channel to the other.

From what you are hearing, and the wave forms; i don’t think you need to have any input to chase the issue. I would use headphones at the amplifier headphone jack (if there is one) and with just the minimal signal path connected and no audio playing turn up the volume and listen just for the noise and try to swap interconnects and speaker cables to look for any changes. You can add in your recording set up to see if the noise you are hearing shows up in the waveforms. That’s how I chase noise.

@surdo "The Sunfire’s speaker outputs produce a low level, but audible on headphones, 60Hz hum (or a harmonic of 60Hz). The Quad, while perhaps more noisy (with hiss) has no 60Hz tone. I tested both the Sunfire’s unbalanced inputs with the Quad 34 preamp and the balanced inputs with the sound coming from a mixing desk. Both produce the hum and the amp produces a hum even with no inputs connected and with the laptop, that’s monitoring the terminals, running on batteries. I’ve tried disconnecting the Sunfire’s earth and reversing the pins, but the hum is unchanged.”

Just went back and read this part again. I don’t see much 60Hz on any of the waveforms, but as I said, there is 120Hz on all of them. Using just the Quad 34 preamp and connected to the 2 amps and no input look for the noise. I am also not clear if your tests were done with 2 different sets of speakers and speaker cables, but to compare the amps, it should be done with the same speakers and speaker cables from either amp. If that is not the case, the speakers and cables cannot be separated from what the amps are doing.

@budjoe Thanks again for your help with this. Yes, the last set of recordings was done with the same cables, speakers (Tannoy), preamp (Quad 34) and recording device. Only the power amps were changed.

I wonder if normalising the peak values is the right way to go. I imagine that the two amps will potentially render peaks quite differently. And if that’s the case, wouldn’t that produce audio mismatched for loudness and with the background noise exaggerated in the recording with the lower peak?

I’ve gone back to my Tannoy/Quad34 recordings and normalised the files, constraining the RMS to -5.0dB. There was no clipping and the two files sound reasonably equal in loudness to me in headphones.

Yes the Quad recording has noise and a thin, grainy sounding hum, but the hum of the Sunfire is louder and deeper, to my ears at least. On the Quad the hum is louder in the right channel. On the Sunfire the hum is louder in the left but more balanced.

All the same...

More important to me than the noise, is the distortion of the piano sound in the Sunfire recording. Are you (or anyone else) able to hear this distortion?

Here are the two files normalised for loudness:

https://escuta.org/webtmp/QuadPiano.flac

https://escuta.org/webtmp/SunfireVoltagePiano.flac

Cheers,

Normalizing should be done to peak amplitude, not RMS. Both of your files have a peak amplitude that is clipped at between 1’08”  and 1’09”. I generally normalize to -1.0 dB of peak. I used values in your original samples to come up with the values I used when initially normalizing them since I wanted to add a little variable as I could to the samples. 

The rest of the values from the waveform indicate the min and max of both the RMS absolute dB are very closely matched.

Haven’t listened to them yet, but noticed that the sonogram for the Quad has some strong tones around 8KHz that is much less in the Sunfire.