Measuring the power of an amp, with an oscilloscope

[HarmonicTHD]

I always use a resistor and pot meter to make a parallel voltage divider across my dummy load. Then I never can short circuit the amp to ground, and do not overload the oscope

A voltage divider will not necessarily prevent you from accidentally shortening your amp, if you have class D or bridged amps to measure. The divider would have to be very high impedance otherwise there is still current flowing from the negative to earth ground. (Differential Probes are afik OP amp based with very very high impedance).

So to be safe: Either use an isolation transformer or a differential probe (see your linked Video).
Or the poor man’s solution, if you can’t afford a differential probe or isolation transformer is to use a two channel oscilloscope and hook up Channel 1+ to Amp Out Positive and Channel 2+ to Amp Out Negative (and disconnect the crocodile clips from your probes altogether so they don’t accidentally touch anything). Then use the math function of the scope to deduct Channel 1 - Channel2.

 

[ajc9988]

I just use a bench power supply with isolation circuit (Riden RD6012P) to float the amplifier boards, but that is if you are hooking up without a power supply that comes with the amplifier. They also have the plugs that can isolate the circuit from creating ground loops.

Other than that, I have an MSO5000 series unlocked oscilloscope. I can get a 1MSa/s 25kHz range FFT with that scope (it is an 8GSa/s scope), equaling about 2 second period. I currently cannot generate the .csv file with the ultrascope software from Rigol, so when I have time, I'll be testing dumping the memory to a USB or saving the samples in some other way, then running my own FFT on the samples.

My question is, would that be sufficient for generating some of the data for THD? THD+N?

Also, I am wondering on the best way to take the FFT data, in dBV, and to properly create those charts seen. I usually have a noise floor around -69dBV with the FFT on the oscilloscope, unless I run it middle of the night with other electronics off and a less noisy mains, allowing for -74dBV up to -80dBV approx. And the MSO5000 allows for up to 15 points on the FFT, where you can vary the excursion, etc. to identify the harmonics and their dBV value.

My issue is making that data usable in charts for the audio community. What I mean by that is I need to find best practices on how to properly setup the input and values to get usable datum, how to convert from the dBV rms values to the charts, I need to build a notch filter to create an easy subtract scenario of the signal carrier and the 1kHz notch filtered to easily perform the ratio of signal noise distortion over noise distortion.

So I would be interested in if a person could point me toward resources, not tell me it is not possible. I can put the caveats of equipment and limitations when I post measurements.

This scope also does Bode plots, which I've had fun looking at to find the gain and phase of the amplifiers. But I'm still relatively new to audio and this is my first oscilloscope, so I'm still learning. Sure, it isn't a spectrum analyzer, but if you pull the points or have a large enough sample FFT, you can have some information that is usable.

So instead of nay-saying, can those with more experience explain how to do it, rather than why not to.

 

[DonH56]

I just use a bench power supply with isolation circuit (Riden RD6012P) to float the amplifier boards, but that is if you are hooking up without a power supply that comes with the amplifier. They also have the plugs that can isolate the circuit from creating ground loops.

Other than that, I have an MSO5000 series unlocked oscilloscope. I can get a 1MSa/s 25kHz range FFT with that scope (it is an 8GSa/s scope), equaling about 2 second period. I currently cannot generate the .csv file with the ultrascope software from Rigol, so when I have time, I'll be testing dumping the memory to a USB or saving the samples in some other way, then running my own FFT on the samples.

My question is, would that be sufficient for generating some of the data for THD? THD+N?

Also, I am wondering on the best way to take the FFT data, in dBV, and to properly create those charts seen. I usually have a noise floor around -69dBV with the FFT on the oscilloscope, unless I run it middle of the night with other electronics off and a less noisy mains, allowing for -74dBV up to -80dBV approx. And the MSO5000 allows for up to 15 points on the FFT, where you can vary the excursion, etc. to identify the harmonics and their dBV value.

My issue is making that data usable in charts for the audio community. What I mean by that is I need to find best practices on how to properly setup the input and values to get usable datum, how to convert from the dBV rms values to the charts, I need to build a notch filter to create an easy subtract scenario of the signal carrier and the 1kHz notch filtered to easily perform the ratio of signal noise distortion over noise distortion.

So I would be interested in if a person could point me toward resources, not tell me it is not possible. I can put the caveats of equipment and limitations when I post measurements.

This scope also does Bode plots, which I've had fun looking at to find the gain and phase of the amplifiers. But I'm still relatively new to audio and this is my first oscilloscope, so I'm still learning. Sure, it isn't a spectrum analyzer, but if you pull the points or have a large enough sample FFT, you can have some information that is usable.

So instead of nay-saying, can those with more experience explain how to do it, rather than why not to.

Rigol MSO5000, not Tektronix MSO5000? Either way, the problem is the 8-bit vertical resolution, which limits the dynamic range for precision amplitude (e.g. THD, noise, etc.) measurements. The FFT resolution and processing (bin) noise floor is not the issue; it's the dynamic range of the front end and digitizer of the 'scope. So you can get useful measurements of some things, just be aware of the limitations of the front end and data conversion (ADC). I am not familiar with Rigol, but many DSOs provide a high-resolution option and the ability to oversample and filter to increase the effective resolution. Oversampling won't help if the front end is only linear to 8 bits or so, however; you'll be able to reduce the noise floor, but not distortion.

The usual solution is to build a notch filter to suppress the fundamental, then you can look down much further into the noise and distortion. There are some good app notes on low-distortion, low-noise op-amp circuits at places like the Analog Devices and Texas Instruments web sites. I don't have any handy, sorry, audio is not my day job.

You can look at Amir's reviews to see what an AP (Audio Precision) analyzer's dashboard set of plots is like. That is a good starting point for presentation since AP units are widely used in the audio world. Distortion is simply dB with respect to the signal level, and for DACs reference is to full-scale output. Most THD+N charts are relative so as long as the reference is defined you should be OK. You might want to use dBW for power amplifiers, reference to power (average watts) instead of voltage, but again most folk will likely focus on how far below the signal the distortion is and not the absolute value of 0 dBxxx.

'Scopes in general, at least the ones I have used through the years, are notoriously finicky about ground connections and frequently do not isolate signal ground very well (even on the $500k models we use at work). Dif probes are a great idea but costly and tend to be somewhat noisy. Using two probes and math in the DSO works well as long as the probes are well-matched (or mismatch is compensated). When I do that, I do not leave probe ground floating (too noisy in my world), but tie them together at a common local ground point. For an amplifier, I probe + and - outputs, and tie both probe ground to the chassis at the same point.

HTH - Don

 

[solderdude]

OK, now I know:
To listen to Miles Davis - doop-bob at the desired volume, i need at least 4v rms per channel
I just attached the oscilloscope while listening.

I see about 23Vpp = 8,2V = 1,4W in 50ohm = 101dB peak SPL

 

[Holmz]

... I am not familiar with Rigol, but many DSOs provide a high-resolution option and the ability to oversample and filter to increase the effective resolution. Oversampling won't help if the front end is only linear to 8 bits or so, however; you'll be able to reduce the noise floor, but not distortion.
…

In general your post is correct.
But oversampling actually does help “a bit”.

Every time we double the sample rate one can thing of that as an extra bit. So we have two samples where we used to have one.
Say a 1 and 2, which average to 1.5
It is just hugely inefficient, so if we go from 58k to 96 we get another bit worth of depth, and 96 to 192k we get a second bit.
We have to go to ~12.8 mega samples/sec to get 8 more bits worth, and it would have been better to just start with a 24 bit ADC versus 16 bit. Or a 16 bit ADC versus 8 bit.

 

[ajc9988]

Rigol MSO5000, not Tektronix MSO5000? Either way, the problem is the 8-bit vertical resolution, which limits the dynamic range for precision amplitude (e.g. THD, noise, etc.) measurements. The FFT resolution and processing (bin) noise floor is not the issue; it's the dynamic range of the front end and digitizer of the 'scope. So you can get useful measurements of some things, just be aware of the limitations of the front end and data conversion (ADC). I am not familiar with Rigol, but many DSOs provide a high-resolution option and the ability to oversample and filter to increase the effective resolution. Oversampling won't help if the front end is only linear to 8 bits or so, however; you'll be able to reduce the noise floor, but not distortion.

The usual solution is to build a notch filter to suppress the fundamental, then you can look down much further into the noise and distortion. There are some good app notes on low-distortion, low-noise op-amp circuits at places like the Analog Devices and Texas Instruments web sites. I don't have any handy, sorry, audio is not my day job.

You can look at Amir's reviews to see what an AP (Audio Precision) analyzer's dashboard set of plots is like. That is a good starting point for presentation since AP units are widely used in the audio world. Distortion is simply dB with respect to the signal level, and for DACs reference is to full-scale output. Most THD+N charts are relative so as long as the reference is defined you should be OK. You might want to use dBW for power amplifiers, reference to power (average watts) instead of voltage, but again most folk will likely focus on how far below the signal the distortion is and not the absolute value of 0 dBxxx.

'Scopes in general, at least the ones I have used through the years, are notoriously finicky about ground connections and frequently do not isolate signal ground very well (even on the $500k models we use at work). Dif probes are a great idea but costly and tend to be somewhat noisy. Using two probes and math in the DSO works well as long as the probes are well-matched (or mismatch is compensated). When I do that, I do not leave probe ground floating (too noisy in my world), but tie them together at a common local ground point. For an amplifier, I probe + and - outputs, and tie both probe ground to the chassis at the same point.

HTH - Don

Yes, the Rigol MSO5000, not the Tektronix MSO5 series. And yes, this is limited to the 8-bit vertical, with up to 12-bit oversampling.

As to the app notes, I've started going through those a bit, but haven't really had the time to sit down and get into it. And I understand the day job comment. I'm in the legal profession. Overclocking computers was my hobby. Then I couldn't get a graphics card, so I took up audio (and wound up spending more setting up an electronics lab and garage shop than if I just spent on an overpriced graphics card).

As to the charts being relative, that is where my question really comes in. For making a relative chart, what is selected for scale? Do you have any papers or sources I should look at to understand better how to go from the data generated to the relative chart?

And yes, I need to get a dif probe, but that is down the road (maybe a christmas present for myself). I've also been doing similar for the common ground point.

Attached is the FFT of a cheap TPA3255 amp (because why not play). I've looked at the AP set of plots. But what I've had a hard time finding is the dB relative scaling and how to calculate that properly (and what points to set for the relative scale). Then, most use A weighting, for better or worse (not going to weigh in on the discussion on A weighting this morning). But I'm open to being pointed toward anything that improves my understanding and teaches me more things. So any and all assistance is appreciated.

Also, thank you for taking the time to respond.

 

[DonH56]

In general your post is correct.
But oversampling actually does help “a bit”.

Every time we double the sample rate one can thing of that as an extra bit. So we have two samples where we used to have one.
Say a 1 and 2, which average to 1.5
It is just hugely inefficient, so if we go from 58k to 96 we get another bit worth of depth, and 96 to 192k we get a second bit.
We have to go to ~12.8 mega samples/sec to get 8 more bits worth, and it would have been better to just start with a 24 bit ADC versus 16 bit. Or a 16 bit ADC versus 8 bit.

Oversampling alone, with a simple filter (no delta or delta-sigma modulation), will reduce the quantization noise floor by 1/2 bit (~3 dB) for every doubling in sampling rate keeping the signal bandwidth the same. E.g., if you filter (limit) the output of a ADC/DAC sampling at 44 kS/s to 22 kHz signal bandwidth , then double the sampling rate to 88 kS/s keeping the filter at 22 kHz, you will pick up 1/2 bit in noise floor (lower noise). That does not affect front-end (analog) distortion, however, and that was my point (perhaps poorly made). If the analog front end is only linear to 8 bits, then oversampling will not improve distortion, just the noise floor.

Delta-sigma converters pick up an additional bit for each order of modulation, and even more when multibit converters are in the delta-sigma loop, but again if the input (or output, for a DAC) analog cells do not provide sufficient linearity then distortion will be limited by that and not the number of bits.

It is probably worth the reminder (not to you, in general) that the simple formula SNR = 6N+1.8 dB for N bits is only for quantization noise and does not include things like circuit noise and distortion.

I have some articles here about sampling and data conversion for conventional and delta-sigma architectures.

 

[DonH56]

Yes, the Rigol MSO5000, not the Tektronix MSO5 series. And yes, this is limited to the 8-bit vertical, with up to 12-bit oversampling.

As to the app notes, I've started going through those a bit, but haven't really had the time to sit down and get into it. And I understand the day job comment. I'm in the legal profession. Overclocking computers was my hobby. Then I couldn't get a graphics card, so I took up audio (and wound up spending more setting up an electronics lab and garage shop than if I just spent on an overpriced graphics card).

As to the charts being relative, that is where my question really comes in. For making a relative chart, what is selected for scale? Do you have any papers or sources I should look at to understand better how to go from the data generated to the relative chart?

And yes, I need to get a dif probe, but that is down the road (maybe a christmas present for myself). I've also been doing similar for the common ground point.

Attached is the FFT of a cheap TPA3255 amp (because why not play). I've looked at the AP set of plots. But what I've had a hard time finding is the dB relative scaling and how to calculate that properly (and what points to set for the relative scale). Then, most use A weighting, for better or worse (not going to weigh in on the discussion on A weighting this morning). But I'm open to being pointed toward anything that improves my understanding and teaches me more things. So any and all assistance is appreciated.

Also, thank you for taking the time to respond.

I guess I do not usually worry about scale since dB itself is a measure of a ratio so is already a relative sort of measurement. It is typical to use 0 dB as full scale for an ADC or DAC since full scale has a specific definition, the largest number the data converter can produce. But 0 dBFS can be 1 V or 10 V (or anything else) depending on the design and implementation. Most people think in watts for power amplifiers so dBW is useful, but you must define the load impedance (resistance, like 8 ohms) to convert from dBV to dBW since your DSO measures voltage and not power (by default; you can use math functions for the conversion, since P = V^2 / R).

But again, my day job is not audio, but signals with much wider (GHz) bandwidth, with a different set of metrics and charts.

 

[Bow_Wazoo]

I just had the following idea:

can I compensate for the low resolution of my oscilloscope, by simply playing the music signal more slowly?
Let's say I reduce the tempo of the music (no problem with Neutron player) by 50%, then the oscilloscope should be able to detect the voltage peaks much easier?!

 

[antcollinet]

I just had the following idea:

can I compensate for the low resolution of my oscilloscope, by simply playing the music signal more slowly?
Let's say I reduce the tempo of the music (no problem with Neutron player) by 50%, then the oscilloscope should be able to detect the voltage peaks much easier?!

If it is bit resolution you are lacking - ie your scope is only 8 bit, then slower won't help (However, 8 bit is probably still enough for a power measurement, it will still be accurate to about +/-0.5%). If it is a sample rate problem (Ie your scope has not enough bandwidth for audio) then it might help. However, you'll now be halving all the frequencies sent to your amp. It will need to have bandwidth down to 10Hz to accurately portray the power level. And your load (speaker) characteristics will not be equivalent at the lower frequencies (unless you are using a resistive load for your measurement)

In any case the power content of music is low at high frequencies, so you'll not lose much accuracy with a limited bandwidth.

What is the reason you are trying to measure power with music instead of test tones?

 

[Bow_Wazoo]

Pure play instinct, to be honest.
I would like to try again to find out, how much power I demand from the amplifier, with my HE-1000V2 Stealth, when I really step on gas (EDM, very loud).
I have chosen a track, and analyzed the spectrum. My peak is at 40hz, TPL -0.1dB. Accordingly, I have prepared a test tone.

In my hought process I forgot that by changing the tempo, I also change the pitch, which I would have to compensate for with pitch control.
Whereby I have no idea how the frequencies in the deep bass change.

 

[HarmonicTHD]

Pure play instinct, to be honest.
I would like to try again to find out, how much power I demand from the amplifier, with my HE-1000V2 Stealth, when I really step on gas (EDM, very loud).
I have chosen a track, and analyzed the spectrum. My peak is at 40hz, TPL -0.1dB. Accordingly, I have prepared a test tone.

In my hought process I forgot that by changing the tempo, I also change the pitch, which I would have to compensate for with pitch control.
Whereby I have no idea how the frequencies in the deep bass change.

Measuring power with for example a sine wave is most demanding for an amplifier. So if your amp passes that, music will be “easy”.

So I would pick individual sine tones eg 20Hz, 100, 1000 etc to 20k and see what happens. Just don’t kill it ;-) (Most commonly 1kHZ is used, mainly as we are quite sensitive in that frequency range). And if you have a multimeter which can measure at those frequencies that would work too aside from the Osci.

 

[Bow_Wazoo]

When I look at my EQ preset, and add that according to the analysis, that my selected track has its peak at about 40hz, most of the energy will take place there.

I just saw that there are now PCI Express cards with 16 bit resolution as oscilloscope. Interesting.

 

[HarmonicTHD]

When I look at my EQ preset, and add that according to the analysis, that my selected track has its peak at about 40hz, most of the energy will take place there.

View attachment 265163

I just saw that there are now PCI Express cards with 16 bit resolution as oscilloscope. Interesting.

Then run a 40hz sine wave into your amp.

In general. What you are trying to do rather requires an FFT and here the usual digital Osci are not resolving enough (even with 16bit). For hobby use rather look into an audio interface (starting around 100USD) or something like the E1DA Cosmos if you want excellent resolution for about 200USD.

But for just measuring power a good multimeter or an 8bit Osci are all what is needed as one usually doesn’t care whether the amp can put out 100w or 100.00001W etc

 

[antcollinet]

Measuring power with for example a sine wave is most demanding for an amplifier. So if your amp passes that, music will be “easy”.

So I would pick individual sine tones eg 20Hz, 100, 1000 etc to 20k and see what happens. Just don’t kill it ;-) (Most commonly 1kHZ is used, mainly as we are quite sensitive in that frequency range). And if you have a multimeter which can measure at those frequencies that would work too aside from the Osci.

Whatever you do though, don't send high frequency tones at full power to a speaker. That will kill your tweeter. I wouldn't go above 1KHz at full power into a speaker, and even then, not for long.

And if you are using a speaker, also use ear defenders.


EDIT : In fact don't use a speaker!

 

[HarmonicTHD]

Whatever you do though, don't send high frequency tones at full power to a speaker. That will kill your tweeter. I wouldn't go above 1KHz at full power into a speaker, and even then, not for long.

And if you are using a speaker, also use ear defenders.


EDIT : In fact don't use a speaker!

Yes. Use a load resistor. (That’s what I use for hobby purposes).

 

[Bow_Wazoo]

8.99 at Amazon.
I am curious.

 

[antcollinet]

8.99 at Amazon.
I am curious.

View attachment 265364View attachment 265365

Worth a go.

They are wirewound though, so may have an inductive characteristic. It would be worth running a frequency response test on them before use.

 

[SIY]

8.99 at Amazon.
I am curious.

Putting aside the fact that this thread is all about using the wrong tool for a simple job, those resistors work fine unless you're trying to get down to the 0.0001% and below world. The issue I found was that the mounting holes needed to be cleaned up a bit so they would make good thermal contact with the heat sink.

 

[restorer-john]

Putting aside the fact that this thread is all about using the wrong tool for a simple job, those resistors work fine unless you're trying to get down to the 0.0001% and below world. The issue I found was that the mounting holes needed to be cleaned up a bit so they would make good thermal contact with the heat sink.

I found those cheap gold anodised Chinese resistors to actually have no appreciable inductance up to 30kHz. I was incredibly surprised considering their low cost. They were better in that respect than my high powered, supposedly 'non-inductive' load resistors I'd been using for years.

Their disipation ratings however are optimistic, but at the price, one can use as many as needed without worry.