Measuring A Modern Discrete Class D Amp

The QA402 can readily measure the performance of modern class D chip amps, even with them surpassing conventional class AB amps. See here and here and here for examples.

But some of the newer Class D amps from companies like Purifi are re-defining what high-performance means from a measurement perspective. Let’s take a look at some of their published specs and see what it might mean if we wanted to measure with at least some margin.

First, we have the following table from their website. The two important pieces from this table is the gain (12.8 dB), the output noise (11.5uV A weighted, which is -99 dBV AW, or perhaps -102 dBV unweighted -96 dBV unweighted. (See @Markyb422 comment below)


Next, let’s build a table from what we know and can estimate from other curves they provide.

In the table above, for 250W output, that means the amp will be generating 31.6Vrms = 30 dBV. We know the gain is 12.8 dB, which means the input to the amp must be 17.2 dBV single ended (30 - 12.8 = 17.2). In volts, this means the driving signal must be 7.24Vrms, which is a lot of drive. If you drive it balanced, then that is 3.6Vrms = 11.1 dBV

Assuming the noise stays constant around -102 dBV (which might be invalid), we can see that the noise appears to dominate the THD+N at lower output levels, and harmonics tend to come up and higher output levels. Roughly, when driving at 250W, we might expect the harmonics to dominate the spectrum (rather than noise) and that the harmonic levels will be around -88 dBV.

With -102 dBV of noise and 12.8 dB of gain, we know the input noise must be below -115 dBV to ensure the noise is dominated by the amp and not the test equipment.

The Balanced Line Transmitter

The QA402, driving balanced at 11.1 dBV per side (17 dBV total) shows harmonic levels about -108 dBc when in loopback. This is seen in the 36 dBV input range. We’re also showing the noise level MINUS the distortion, which is -78.6 dBV. So, while the harmonic level is fine, the noise level is too high at this setting to drive this amp. BUT, we don’t know if the noise is due to the QA402 generator output OR the input. Presumably, it’s the input side and the result of the attenuator level needed to hit the THD performance desired.

Now, let’s run the same out through the notch on the QA480. This will knock the 1 kHz down by 60 dB or so, and that in turn will let us use a more sensitive range on the QA402 input. That plot is below. Note the 2H is roughly in the same place–a little better-- (about -95 versus -92–this 3dB came from the notch performance at 2 kHz). But with the 0 dBV input range on the QA402, we can get a real feel for the noise out of the DAC. And we see with an 11 dBV output, the DAC noise is around -90 dBV.

Remember that to measure 1W out of the Purifi amp, we’d need our noise + distortion to be somewhere around -98. And clearly, we’re not there. But fear not: Measurement is all about figuring out what combination of things can get you where you need to go.

What if we ran the output single-ended into a single-stage bandpass filter with the response shown below. This accomplishes two things: First, it dramatically lowers the 20 to 20 kHz noise level. Second, it should knock down the second harmonic assuming the filter itself doesn’t contribute anything.


A circuit was built with careful attention paid to the noise. A SPICE sweep of the notch is shown above, and the measured circuit is shown below. Note the 20 dB attenuation at 2 kHz. But more importantly, note that half the spectrum, from 10 kHz to 20 kHz, has been attenuated by some 40 dB. This will help our noise enormously.

The lineup now is QA401 output → Bandpass Filter Input → QA480 Notch Input → QA480 Notch Output → QA402 Input. The plot from that lineup is shown below:

The first thing to note is the shape of the noise is much flatter. The fact that we could see the notch in the first notch plot means that the noise level of the QA402 output was higher than the QA480 notch self noise, and the QA480 was “shaping” the noise.

But in this plot we can’t see the QA480 shape any longer, which means we’re below the QA480 self-noise. Also, now look at our noise minus distortion: almost -104. That is about a 15 dB win. Plus, the QA480 notch is likely now contributing some noise. We’ll figure that out later.

But in short, we’ve got a path forward on the output side of things.

The Balanced Line Receiver

So what would we want/need on the receiver side of things? First, we want the ability to connect differentially with some attenuation directly across a load. Going back to the table below, let’s aim to hit up to 250W without needing the attenuator an attenuator we’re going to build into the Balanced Line Receiver. That means at 250W we’d expect the distortion peaks from the Purify to hit around -90 or so, and the noise (minus the distortion) should remain around -98 dBv.

So, let’s start with a instrumentation opamp, followed by a 2nd order low pass filter with a corner at 67 kHz, and then a super low noise notch. By “low noise” I mean we’ll take the QA480 notch, bump the “tank” caps up by 10X and drop the R by 10X. And the follow all that with an 18 dB gain stage.

A sweep of the receiver side of things looks as follows. Note we have roughly 55 dB of attenuation, and 18 dB of gain overall.

Next lets short the input of the line receiver and measure the noise:

Alright! This is looking good: This is -98.5 dB of noise with the 18 dB of gain. Input referred, this means our noise at the input is -117 dBV. Remember the amp output noise was -102 dBV (unweighted), and the -117 dBV figure gives us plenty of margin.


And how does this setup look in loopback? That is, the QA402 DAC → Line Transmitter → Line Receiver → QA402? That plot is below:

Remember, what we’re seeing here isn’t as things actually are. We put 11.1 dBV in. We know the bandpass didn’t impact that signal level. But the notch did. If it weren’t for the notch, we’d expect to see 11.1 dBV out plus another 18 dB of gain that was built into the line receiver. Everything outside of the notch has already seen the 18 dB of gain.

So let’s correct by creating a user filter that bumps the center freq from -15.6 to the expected level of 11.1 dBV plus the 18 dB of gain, or 29.1 dB. That is, we’d expect the output level to be 29.1 dBV, but due to the notch it is just -15.6. Overall, we need to add 29.1 + 15.6 = 44.7 dB dB to the fundamental, and we know the notch knocked down the 2H by 9 dB or so, we can compensate for that too.

The user weighting function thus becomes:

0, 0
990, 0
991, -44.7
1009, -44.7
1010, 0

1990, 0
1991, -9
2009, -9
2010, 0

24000, 0

And once that is loaded the plot becomes:

And now our THD+N is about -118 dB in loopback. And note that it is dominated by harmonics, likely from the line transmitter side. But with some straightforward additional analog processing, we’re within striking distance. Another 10 dB on the transmit side harmonics and a bit more on the noise would yield some amazing performance. For now, we’re not quite there yet, but over the coming months we’ll spend some time going through the design in more detail, block diagrams, schematics and simulations, and see how much better we can do.

Stay tuned, and feel free to ask questions or offer observations and/or corrections.

Thanks, Matt


Revision C of the hardware discussed above was built. This added a pushbutton to enable/disable the 18 dB attenuator on the Line Receiver side, and also changed the bandpass on the Line Transmitter side to be second order instead of first. As we’ll see below, that didn’t yield any win.

The block diagram is shown below:

Remember the aim here is to develop an external box that can let the QA402 measure at 1 kHz a THD+N in excess of -121 dB or so, but ideally -125 dB.

Characterizing the blocks on the new hardware is shown below:

Line Transmitter Bandpass Filter

A sweep of the bandpass filter shows the expected 40 dB suppression materialized. The hope here was to suppress the harmonics that had appeared in this stage.

Let’s take a look a the QA402 output at 0 dBV, running through the QA480 notch:

The notch has 18 dB of gain post-notch. Note we’re seeing 2H at -135, and that is after the 18 dB of gain, but also includes 9 dB of atten from the notch. So, going into the notch we have our fundamental at 0 dBV and our 2H at -150 dB. In other words, the QA402 performance here in terms of harmonics is exceptional…it’s the noise that is hurting.

Running the QA402 into the Line Transmitter, and then running that output to the QA480 notch and then taking that into QA402 give the plot below:

This is mostly unchanged from the REVB hardware. We should have seen an extra 20 dB of 2H suppression, but it didn’t materialize, which suggests the topology and/or caps used in the bandpass are adding distortion. Other NP0 caps will tried. But in short, for REVD we’ll go back to a single stage look for some wins via cap selection (for harmonics) and resistor values (for noise) as SPICE suggests some more is to be had via resistor values.

Line Receiver Frequency Response

The overall response of the receiver is shown below. The notch can be clearly seen as can the roll-off around 40 kHz

With both receiver inputs shorted, the output noise is about -98 or -99 dBV. This gives us an input-referred noise around -118 dBV.

Balanced Loopback

Going back to this table:

Focusing on the 250W case, we want a drive level of 17.2 dBV, which would be 11.2 dBV balanced. Absent the DUT amplifier gain, we’d expect the output of the full loopback (through transmitter and receiver) to be 11.2 dBV plus the 18 dBV gain from the receiver stage. Thus, 29.2 dBV at the output.

The plot below shows the loopback. The 2H and 3H are determining the THD+N. The noise floor is -93.8, which is about -122.5 dB below the peak.

So, a little progress was made in REVC though not much. We need to pull out a few more dB in noise from the TX side and pull another ~10 dB out in the harmonics.

More coming…

1 Like

Hi Matt,
I’m a newby and pretty new to forums. I bought a QA401 and QA480 a few months back and have read and re-read your write-up on “Measuring A Modern Discrete Class D Amp”, hoping to build my understanding.
I re-read this again tonight and was struck by your third paragraph:

I thought that the A-weighting rolled off the energy at low and high frequencies to mimic the effect of human hearing, so with the Purifi specced at 11.5uV A weighted (-99dBV AW) it would be NOISIER unweighted so perhaps -96dBV unweighted versus your -102dBV unweighted. I hope I am correct as this makes your task about 6dB less onerous, no?

Hi @Markyb422, good catch, you are indeed correct! I’ll correct it

Just to add to the amendment you made:

Change these two mentions of -102dBV to ‘-96dBV’ and “input noise must be below -109dBV to ensure the noise is dominated by the amp and not the test equipment.”

So the need for a test equipment noise floor at -109dBV not -115dBV gives a 6dB easing in its design spec.


-96dBV (unweighted) against the input noise of -117dBV you achieved means even more margin; enough to measure a SOTA Benchmark AHB-2 power amp?

1 Like