Verify Power Amplifiers Distortion with FM Modulation Technique

[pma]

Verify Power Amplifiers Distortion with FM Modulation Technique

One of the possible methods to test amplifiers at multiple frequencies is a frequency modulation, as an alternative to multitone testing. There is a significant difference to the multitone signal – FM modulated signal keeps the full amplitude swing and thus stresses the amplifiers under test at much higher power than the multitone test. The amplifiers were tested into 4ohm resistor load.

1. The test signal

Over the years I have developed a set of FM modulated test signals, one of them is the FM_1kHz_900Hz_200Hz signal (1kHz carrier modulated by 900Hz with modulation frequency 200Hz). The signal looks like this, in time domain:

and the frequency spectrum like this (in logarithmic scale of frequency axis)

and like this (in linear scale of frequency axis)

The spectral lines of the test signal are placed between 200Hz and 4kHz with 200Hz displacement. Above 4kHz, all the spectral lines are below -150dBr.

I will test 2 different amplifiers with this signal, after having the DAC-ADC system tested as well. Two sets of measurements will be collected, one with a logarithmic scale of frequency axis and another with linear scale of frequency axis. It is then easier to note certain differences.

2. Plots in logarithmic frequency scale

2.1. Measurement of DAC-ADC chain, Topping D10s – E1DA Cosmos ADC

Above 4kHz, all the spectral lines are below -130dBr. It is very good, however 20dB worsening compared to the test signal itself. Noise modulation can be seen below 4kHz.

2.2. Measurement of PMA NC252MP amplifier

Above 4kHz up to 20kHz, all the spectral lines are below -106dBr (related to signal components). Above 20kHz, we can see class D and SMPS mess.

2.3. Measurement of PM-AB2 class AB MOSFET amplifier

Above 4kHz (up to 40kHz limit), all the spectral lines are below -104dBr (related to signal components). There is no added ultrasonic mess, as it is the linear amplifier.

3. Plots in linear frequency scale

3.1. Measurement of DAC-ADC chain, Topping D10s – E1DA Cosmos ADC

3.2. Measurement of PMA NC252MP amplifier

3.3. Measurement of PM-AB2 class AB MOSFET amplifier

4. Conclusion

FM modulation may be an alternative to IMD tests and multitone tests. It stresses the amplifier under test by a full signal amplitude, thus revealing large signal non-linearity. The amplifiers under test have both behaved very well, accordingly to their “traditional” measurements. Distinctions between the class D and the linear amplifier can be clearly seen, especially with the linear frequency axis.

 

[KSTR]

Very nice.

Slightly off-topic but you and others might find this useful.

Your test signal (which still is a multione, conceptually) is somewhat similar to a general periodic test signal class I came to like and what I'm calling a Newman Sweep.
The Newman Sweep basically is a multitone with (typically) the smallest possible crest factor. This is achieved with a special phase relationship of the components found by Mr.Newman (see attached paper) which in the end yields an "endlessly overlapping sweep".
Newmans Sweeps can be generated for arbitrary sets of test tones and any sequence length which makes them very versatile, they can be used with time-domain averaging and don't require FFT windowing.

Q&D example of a signal similar to yours:

Note I've used slightly non-harmonic multiples here, that's why the sub-sections don't look exactly the same (but do repeat every 16384 samples).

Spectrum:

Crest factor might be not as low as for your signal but is quite good with only 2.6dB.

I have a crude C source code to generate the sweeps (and other multitones) but it is not yet ready for general use -- unless you're a programmer...

 

[MaxwellsEq]

Fascinating! I've never seen either approach before.

 

[pkane]

Verify Power Amplifiers Distortion with FM Modulation Technique

One of the possible methods to test amplifiers at multiple frequencies is a frequency modulation, as an alternative to multitone testing. There is a significant difference to the multitone signal – FM modulated signal keeps the full amplitude swing and thus stresses the amplifiers under test at much higher power than the multitone test. The amplifiers were tested into 4ohm resistor load.

1. The test signal

Over the years I have developed a set of FM modulated test signals, one of them is the FM_1kHz_900Hz_200Hz signal (1kHz carrier modulated by 900Hz with modulation frequency 200Hz). The signal looks like this, in time domain:

View attachment 232462

and the frequency spectrum like this (in logarithmic scale of frequency axis)

View attachment 232463

and like this (in linear scale of frequency axis)

View attachment 232464

The spectral lines of the test signal are placed between 200Hz and 4kHz with 200Hz displacement. Above 4kHz, all the spectral lines are below -150dBr.

I will test 2 different amplifiers with this signal, after having the DAC-ADC system tested as well. Two sets of measurements will be collected, one with a logarithmic scale of frequency axis and another with linear scale of frequency axis. It is then easier to note certain differences.

2. Plots in logarithmic frequency scale

2.1. Measurement of DAC-ADC chain, Topping D10s – E1DA Cosmos ADC

View attachment 232465

Above 4kHz, all the spectral lines are below -130dBr. It is very good, however 20dB worsening compared to the test signal itself. Noise modulation can be seen below 4kHz.

2.2. Measurement of PMA NC252MP amplifier

View attachment 232466

Above 4kHz up to 20kHz, all the spectral lines are below -106dBr (related to signal components). Above 20kHz, we can see class D and SMPS mess.

2.3. Measurement of PM-AB2 class AB MOSFET amplifier

View attachment 232467

Above 4kHz (up to 40kHz limit), all the spectral lines are below -104dBr (related to signal components). There is no added ultrasonic mess, as it is the linear amplifier.

3. Plots in linear frequency scale

3.1. Measurement of DAC-ADC chain, Topping D10s – E1DA Cosmos ADC

View attachment 232468

3.2. Measurement of PMA NC252MP amplifier

View attachment 232469

3.3. Measurement of PM-AB2 class AB MOSFET amplifier
View attachment 232470

4. Conclusion

FM modulation may be an alternative to IMD tests and multitone tests. It stresses the amplifier under test by a full signal amplitude, thus revealing large signal non-linearity. The amplifiers under test have both behaved very well, accordingly to their “traditional” measurements. Distinctions between the class D and the linear amplifier can be clearly seen, especially with the linear frequency axis.

Looks very good, Pavel! Am I understanding this correctly, you used a 1kHz tone that was FM-modulated by 200Hz with 900Hz deviation?

 

[DonH56]

The test signal is a 1 kHz tone with 900 Hz spread (frequency deviation) at 200 Hz (spreading) rate? Is that 900 Hz p-p or +/-900 Hz?

Peak power levels should (can) be the same for multitone (or IMD) test tones and FM sweeps (which could be considered another flavor of multitone since FM sweeps frequency, but by convention "multitone" implies fixed frequencies, so I agree in treating the differently). For multitone testing, the individual tones are decreased in amplitude so that, when they are all in phase ("line up"), the peak signal is at 0 dB referenced to full scale. How often that happens depends upon how the tones are phased, how many tones are used, etc. An FM sweep may provide longer "dwell" time at the peaks and thus stress the system more; the first set of plots show that pretty clearly.

I have performed FM sweeps (and chirps) many times in the RF (RADAR) world but never thought about applying them to audio (outside the usual FM tuner tests, natch). It would be interesting to see a comparison among conventional single-tone sweeps, two-tone (IMD) sweeps, FM sweeps, and multitone tests for noise and distortion. I have in the past performed NPR (noise power ratio) tests on audio circuits but the idea never caught on (it is a common RF system test that was originally developed for analog FDM -- frequency division multiplexed -- telephone circuits). My gut says FM sweeps have the same potential to catch the occasional unexpectedly poorly-performing circuits as multitone and NPR tests. Most designs behave classically IME, but strange things can and do happen that show up in various multitone tests and not caught by single-tone tests.

Neat stuff - Don

Edit: Others said the same thing; I started this post, then had to divert to work stuff before coming back to it, so am late to the party.

 

[pma]

I am adding the results of AIYIMA A07.

Please note the highest noise, noise floor of the 3 amps under test. Please also note the highest distortion, the highest content of modulation products spread up to 40 kHz measuring range. On the other hand, very good suppression of power supply residuals.

 

[pkane]

Added FM to test signal list. Does this look right? Crest factor is indeed, excellent.

 

[JohnPM]

Your test signal (which still is a multione, conceptually) is somewhat similar to a general periodic test signal class I came to like and what I'm calling a Newman Sweep.
The Newman Sweep basically is a multitone with (typically) the smallest possible crest factor. This is achieved with a special phase relationship of the components found by Mr.Newman (see attached paper) which in the end yields an "endlessly overlapping sweep".
Newmans Sweeps can be generated for arbitrary sets of test tones and any sequence length which makes them very versatile, they can be used with time-domain averaging and don't require FFT windowing.

Crest factor might be not as low as for your signal but is quite good with only 2.6dB.

REW's linearly-spaced multitones start with Newman phases before optimising to reduce crest factor. Crest factor of your signal is probably 5.6 dB rather than 2.6, I guess you use the "FS sine rms is 0 dBFS" option, i.e. the dBFS rms per AES, which isn't really the rms of the signal . Hard to get a multitone CF below the 3 dB of a sine wave.

 

[KSTR]

REW's linearly-spaced multitones start with Newman phases before optimising to reduce crest factor. Crest factor of your signal is probably 5.6 dB rather than 2.6, I guess you use the "FS sine rms is 0 dBFS" option, i.e. the dBFS rms per AES, which isn't really the rms of the signal . Hard to get a multitone CF below the 3 dB of a sine wave.

Yes, I tend to use relative crest factor (vs the sine crest factor) as this is much more intuitive to me.

May I ask how you optimize your signal, is this a brute force approach?

 

[pma]

Added FM to test signal list. Does this look right? Crest factor is indeed, excellent.

Hi Paul, thank you. It looks right. I get this distribution, peak rms almost equals to average rms

The test signal is a 1 kHz tone with 900 Hz spread (frequency deviation) at 200 Hz (spreading) rate? Is that 900 Hz p-p or +/-900 Hz?

1KHz carrier, 900Hz delta F modulated by 200Hz. Index 4.5.

The Newman Sweep basically is a multitone with (typically) the smallest possible crest factor.

My test signal is not a multitone, thus the "crest factor" is maximized.

Looks very good, Pavel! Am I understanding this correctly, you used a 1kHz tone that was FM-modulated by 200Hz with 900Hz deviation?

Exactly! Of course you may play with another delta F/Fmod/Fc ratio, to get wider or narrower modulation.

------------------------------------------------------------

I forgot to mention the power used for the tests. It differs according to the amp's max. power.
NC252MP: 250W rated power, tested at 48W
PM-AB2: 35Wrated power, tested at 10W
AIYIMA A07 (27V PSU): 70W rated power, tested at 19W

BTW, this kind of test shows one of the reasons why high order harmonics in the spectrum are to be avoided, by design. High order harmonics create (with a more complex signal) a wide-spread hash of grass that apparently rises the noise floor, in fact reducing the usable dynamic range.

 

[KSTR]

My test signal is not a multitone, thus the "crest factor" is maximized.

It may not be so easy to generate your signal with our typical multitone algorithms but the result shows a (small) set of fixed discrete frequencies which can thus be generated by a bunch of oscillators at those frequencies and levels and the specific (start)phases, which is the definition of a multitone, isn't it?
I agree that with this definition basically almost everthing is a multitone which doesn't help us that much in practice

 

[MaxwellsEq]

BTW, this kind of test shows one of the reasons why high order harmonics in the spectrum are to be avoided, by design. High order harmonics create (with a more complex signal) a wide-spread hash of grass that apparently rises the noise floor, in fact reducing the usable dynamic range.

This has always seemed intuitively the case to me, but this looks like a valid and repeatable visual test for it.

 

[JohnPM]

May I ask how you optimize your signal, is this a brute force approach?

I use a slight variation of the process set out by Yang et al.

 

[pma]

Test signal simulation - the result is in accordance with the real test signal used.

 

[pkane]

It may not be so easy to generate your signal with our typical multitone algorithms but the result shows a (small) set of fixed discrete frequencies which can thus be generated by a bunch of oscillators at those frequencies and levels and the specific (start)phases, which is the definition of a multitone, isn't it?
I agree that with this definition basically almost everthing is a multitone which doesn't help us that much in practice

If you want a really low crest-factor test signal, try a square wave. It's a linearly spaced multitone with decreasing amplitudes

 

[pma]

Regarding the spectrum of the test signal, wideband FM modulation contains infinite number of spectral lines at frequencies Fc, Fc ± Fm, Fc ± 2Fm, Fc ± 3Fm ........, and so on. Amplitudes of spectral components are defined by Bessel functions at the corresponding modulation index ß, ß = dF/Fm. dF is the frequency deviation and Fm is the modulating frequency. We have ß = 4.5, so the red line shows where to read amplitudes of spectral components. Jo(ß) is for Fc, J1(ß) is for Fc ± Fm etc.

The selected index 4.5 makes similar amplitudes of the first 5 spectral components. Our spectrum ends at 200Hz at the low frequency side, as in our real life it is difficult to handle negative frequencies .

The end of the school lecture .

 

[DonH56]

The end of the school lecture .

Sorry, I shall try harder to not post in your threads.

 

[BR52]

Thank you all, it's so much fun to see in this thread here enthusiastic people driving audio measurement with new ideas. Inovation needs discussions please don't stop.

 

[pma]

Sorry, I shall try harder to not post in your threads.

This is really not necessary to see it this way. I am self assured that complementary approach of a set of members is needed to get the complete view. Moreover, it was not meant personally, but rather as a joke. I am not a native speaker, so I am sorry if that sounded offensive.

 

[pma]

Thank you all, it's so much fun to see in this thread here enthusiastic people driving audio measurement with new ideas. Inovation needs discussions please don't stop.

"Think different" is my slogan as well. You know, I was born in 1955 .