This post explains how to measure amplifier output noise using the QA401 Audio Analyzer and the QA451 Programmable Load.
Amplifier output noise is generally expressed as an RMS voltage reading in a given bandwidth. Since most finished amp makers are pretty loose with their specs, let’s turn to the silicon vendors as a first point of reference. Below is the spec for the TI TPA3255:
Notice they have specified 85 uV, A weighted, with an AES17 filter and the amplifier input grounded. We can convert this figure to dBV: 20 * Log10(85uV) = -81.4 dBV. The AES filter they have specified is effectively a brickwall filter that kicks in just beyond 20 kHz. So, we’ll treat this figure as a 20 kHz bandwidth (Remember: bandwidth must always be discussed when talking about noise!). The TPA3255 peaks at 260W (4 ohms, 1% THD+N), which is 32.2Vrms. So, the max sine signal out of the amp is 32.2V rms, and the noise is 85uV rms. This gives a best case signal to noise of 20*Log10(32.2/85u) = 111.7 dB.
If you are playing the TPA into 4 ohm speakers rated at 95 dBSPL per watt, then that means the amp output needed to deliver 1W would be 2Vrms, and thus the noise would be 20*Log10(85u/2) = 87.4 dB below the 1W tone. Which would be 87.4 dB below the 95 dBSPL. Thus, the noise would be 95 - 87.4 = 7.6 dBSPL at 1m. If your listening room has an ambient noise floor of 35 dBSPL (which would be phenomenal), then your noise at 1m is far below that and thus cannot be heard. At all. If you stick your ear right up to the speaker, you might increase the noise level at your ear by 30 dB, so at that point it would be possible to hear the noise in a very, very quiet room.
Now, is the TPA3255 the quietest high-power Class D amp out there? Not even close. The Purify 1ET400 claims 11.5uV.
Putting these two amps into a table gives the following (41.2Vrms corresponds to 420W into 4 ohms):
Because the Purifi has much more output AND much lower noise, we see the best-case SNR jump a staggering 20 dB on the Purifi. You can also see that using the same speakers we used on the TPA3255 would yield a noise floor below 0 dBSPL. Since 0 dBSPL is the threshold hearing in a world without any noise, this indicates this amp noise cannot be heard even if you were in a perfectly quiet world. If you pressed your ear right up to the speaker, you’d still not be able to hear it unless you were in one of the quietest rooms in the world.
Quite an achievement.
Measuring Amp Noise
The setup below can be used to measure amplifier noise. Using the QA451, we’ll measure the amp noise using a single-ended connection to the QA401. Note the negative inputs on the QA401 (L- and R-) are both shorted. We’ve also shorted the input to the amplifier.
Connect the amp as shown to the QA451, and the QA451 to the QA401.
And then, crank the amp as the noise measurement should be made at max gain. Don’t set the amp to max at first. As you validate your setup and gain confidence in the numbers, you can begin to raise the amp gain. But initially, make sure nothing is shorted and that everything seems happy.
Once you are confident in your setup, then unplug the load connections on the front of the QA451. This will let us look at the self-noise of the QA451.
With the 4-conductor QA451 load inputs disconnected, set the QA451 load to 4 ohms. This will ensure a low-impedance is present on the filter inputs of the QA451 (see more info on the QA451 internal architecture here):
Now, on the QA401, select the File->Reset Startup Defaults (to ensure a common starting point), select the +6 dBV input level and start running. Note the measured RMS, from 20 to 20 kHz, is -106.5 dBV. This is a full range noise measurement without any weighting.
The QA451 has 12 dB of attenuation built-in, and we haven’t accounted for that yet. So let’s indicate that right now. By control+clicking on the dBV button, we can specify any input or output offsets. 12 dB of attenuation is the same as -12 dB of gain:
And with that specified, now let’s take a look at the main plot again. Note that two things have changed. First, we see an on-screen message (ExtGain) that signifies we have a non-zero value for the input and/or output gain. Note, too, that the RMS has increase by 12 dB–from -106.5 to -94.4. In other words, the QA401 has adjusted the reading to show us the INPUT level to the QA451. Without this adjustment, we’re looking at the OUTPUT level of the QA451. But we want the INPUT to the QA451 because this is the same as the amp output. This is known as an “input referred” measurement.
The RMS value of -94.4 dBV can be converted to volts using 10^(-94.4/20) = 19.1 uV. Or you could set the option in the QA401 to do this for you automatically.
Now, there’s one more button to press. Since noise measurements are made with A-weighting specified, we can press that now:
And the resulting plot becomes as shown. Note the low-end and high-end have been rolled off. This is the A-weighting filter at work. It’s also dropped our RMS noise from -94.4 to about -97 dBV–about 2.5 dB. And -97 dBV is 10^(-97/20) = 14.1uV when expressed linearly.
Connecting the TPA3255, which is running from a SMPS designed for industrial use and set to 51V, we get the following. For this plot, I’ve switched the RMS reading from log to linear, so we can more readily compare with the datasheet:
Note in the above plot that the RMS noise is measured at 91.8uV, while TI specs the typical noise at 85 uV. Undoubtedly TI is using a very expensive low-noise lab supply to power the TPA3255 for their measurements. In the plot above, we’re not. I don’t know if that accounts for the very small difference or not. But in short, the agreement is very good with TI’s published figures.
Good enough for Purifi?
Is the setup above good enough to measure the state-of-the-art Purifi amp? We’ve measured the setup noise floor at 14.3uV, and the Purify amp is 11.5 uV. What we know is that if we combined two equal noise sources, we’d see the combined noise rise by 3 dB. Here, they are almost identical, and if we combined them mathematically we’d get 18.3uVrms (sqrt(14.3^2 + 11.5^2)).
So, you could confirm the performance of the Purifi pretty quickly using the setup above, and you’d see the noise floor inch up just a bit when the Purifi was connected. But you couldn’t be certain if it was much better than the 14.3uV noise floor. In other words, the Purifi amp could be 1uV of noise or 10uV of noise, and we’d still measure about the same. If it were 20 uV, we’d definitely be able to see that. So, the best we could say right now is that the Purifi amp is meeting it’s spec OR if the Purifi amp was failing to meet its spec.
If you need to know with 100% certainty, then the solution is pretty easy: Follow the Purify amp output with 30 dB of low-noise gain (the QA470 has a general purpose low-noise amp) and the resulting setup would let you measure the Purifi amp with probably 20 dB of margin.