• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Effect of measuring bandwidth on class D amplifier measured THD and THD+N parameters

pma

Major Contributor
Joined
Feb 23, 2019
Messages
4,777
Likes
11,174
Location
Prague
Effect of measuring bandwidth on class D amplifier measured THD and THD+N parameters

It has been quite long lasting discussion here on measuring bandwidth for THD+N plots vs. amplifier power, measured at various frequencies. Especially for class D amplifiers, THD+N results get rapidly worse, at higher frequencies (>4kHz), when we use higher measuring bandwidth (BW) than 22kHz. 22KHz allows only for 3rd harmonics of 7kHz sine tone, which is insufficient to estimate amplifier's non-linearity. It was usual, in the days of linear amplifiers, to use much wider BW, at least 80kHz, when measuring THD and THD+N. My intent is not to start another pointless discussion if 22kHz BW is enough to measure amplifiers or not, as such discussions become emotional and some commercial interests are hidden under the surface as well. My intention is to show how wider bandwidth is reflected in THD and THD+N vs. power measurements.

As a DUT, my PMA-NC252MP amplifier built with Hypex NC252MP module is used. The amplifier is in class I safety class device, with 3-wire mains cord using L, N and PE conductors. Analog signal ground is connected to chassis at each input XLR connector (pin 1 to chassis). The amplifier is not in the BTL mode. OUT-Hi terminal is amplified signal, OUT-Lo terminal is internally connected with signal ground in the module PCB. The module contains built-in SMPS power supply. All this information is important.

1. Test setup

The test circuit is in the following image. The DUT amplifier is driven from Topping D10s DAC via SE-to-balanced converter, to the amplifier balanced input. This converter must be used to get the best S/N results. 3-wire RCA-to-XLR cable is not enough as it does not fix impedance imbalance. The DAC is connected to the PC via USB-ISO isolated USB link. This is important to cut the signal ground loop. Output of the DUT amplifier is connected by means of a 3-wire method to the balanced RC lowpass filter, resistive divider and then to the input of E1DA Cosmos ADC, which is directly connected to the PC via USB link. The PC is supplied from the class II AC/DC supply adapter, so its ground is not directly connected with mains PE or PEN conductor.
isol trafo test setup.png

The DUT amplifier is supplied from 230V/50Hz mains via an isolation transformer with 120pF primary/secondary capacitance. This is again important, especially for SMPS supplied amplifiers. The eefect of this isolation transformer will be shown later, its main purpose is to minimize effect of SMPS leakage current.

2. Measurements of THD and THD+N vs. power with 22kHz and 45kHz bandwidth

Measuring bandwidth of 22kHz and 45kHz was used. The reason of 45kHz limit is that the Topping D10s with USB-ISO adapter does not work with higher sampling frequency than 96kHz. The operation without USB-ISO was a subject to additional errors induced by circulating loop currents, so such option was rejected and the DAC was always connected via USB-ISO. The 45kHz BW still allows to evaluate 5th harmonic of the 9kHz sine wave.

The RC lowpass filter, that is necessary to prevent ADC input from high dv/dt induced errors, has some effect in the measuring band. The filter was used in all cases, frequency response of the NC252MP with this filter is shown below. At 20kHz, we can see attenuation of -2.5dB, at 40kHz it makes -7.1dB.

NC252MP FR with lowpass sm.png


Measurements of THD+N vs. power at various frequencies

PMA-NC252MP mains with isolation trafo, load 4ohm, thdnpower.png


The plots for 22Hz, 1kHz, 4kHz and 6kHz are measured with 22kHz BW. In case of 6kHz measurement, the highest harmonic measured is only 3rd, H3.
The plots for 10kHz and 14kHz are measured with 45kHz BW (but let's not forget some effect of the RC low pass filter), the highest harmonic of the 14kHz sine is again 3rd, H3.

The THD+N plots are prone to mix distortion and noise effects, so another measurement of “pure” THD, without “N” noise component, was done and is shown below.

Measurements of THD vs. power at various frequencies

PMA-NC252MP mains with isolation trafo, load 4ohm, thdpower.png


The plots for 22Hz, 1kHz, 4kHz and 6kHz are measured with 22kHz BW. In case of 6kHz measurement, the highest harmonic measured is only 3rd, H3.
The plots for 10kHz and 14kHz are measured with 45kHz BW (but let's not forget some effect of the RC low pass filter), the highest harmonic of the 14kHz sine is again 3rd, H3.

We can see now that it is not only noise that makes the results with higher BW (45kHz) worse, but there is a considerable and inevitable worsening of results due to distortion harmonics that appear above 22kHz.

The effect of isolation transformer in the DUT power supply line

The following spectrum measurements show 100W/4ohm/1kHz spectrum of THD and THD+N, once measured with amplifier directly connected to mains and then via the isolation transformer. The effect of the isolation transformer is indisputable. The main reson of isolation transformer use is to eliminate capacitive leakage currents from SMPS. It is pointless to measure mains voltage “distortion” with and without the isolation transformer, as it is not the reason why this component is used, to improve whole chain S/N.

PMA-NC252MP mains without isolation trafo, load 4ohm.png


PMA-NC252MP mains with isolation trafo, load 4ohm, BW22kHz.png


Please note that the area above 1kHz, with isolation transformer, is now free of usual spurious, SMPS leakage currents induced mess, that we often see in NCore MP modules measurements. Please also an improvement in THD+N and lower amplitude of ultrasound garbage with the isolation transformer. The area below 1kHz now contains only the mains lines resulting from the all-in-one concept (SMPS+amp shared on mutual board).
 
More detailed view on isolation transformer effect

Spectrum without isolation trafo
NC252MP_100W4R_noisoltrafo.jpg

Please note spurious spectral lines with equidistant spacing of 100Hz through whole spectrum


Spectrum with isolation trafo
NC252MP_100W4R_isoltrafo.jpg


Please also note that measurements are made in a real configuration of audio chain, as used for listening purposes.
 
Those don't seem to be that bad, when using static sine wave measuring.
Try two sine waves in the same time see what happens to IMD (or more relevant, the third-order intercept - TOI/IP3).

Then remember that musical programs have an infinity of sine wavers (Fourier transformation).

PS: I know I will be getting a lot of hate from class D fans, it happen before on this forum.
 
I haven't heard that many class-D amplifiers... I have a class-D car amp and I assume my AVR is class-D.

But, I've never heard distortion from ANY amp that wasn't broken or overdriven. So it's not something I even think about. And I rarely hear frequency response problems... I've heard a few old tube amps with rolled-of highs and maybe weak bass or boosted mid-bass.

Sometimes there is audible noise so that's the only thing I'm concerned with. My previous AVR has noise that I could hear when the room was quiet. I don't remember if it was hiss or whine. And IIRC it was ground-loop noise that went-away when I disconnected the cable TV connection so it wouldn't have shown-up in the specs or lab measurements. I've also had a soundcard with audible noise when the hard drive was accessed.
 
Let's debunk some myths on class D amplifier measurements, namely that measurement bandwidth should not be wider than 22kHz, anything above is unimportant, ... bla bla ...

As an example I am showing measurements of a stone-age class D amplifier, UcD180, in my implementation described here:


I have just re-measured it, using the Topping D10s DAC and E1DA Cosmos ADC (post #1), with measurement bandwidth of 45kHz, load 4ohm, test frequencies from 22Hz to 14kHz.

This amplifier, though its distortion parameters are not SOTA, performs remarkably consistent regardless the higher measuring BW=45kHz. And I did not have to use the mains isolation transformer (as I did in case of NC252MP), because my implementation use a traditional linear power supply wit toroidal transformer, diode bridges and capacitor bank.

So here we go, first THD vs. power.
UcD180_thdpower_4ohm_22-14kHz_BW45kHz.png



Next, THD+N vs. power
UcD180_thdnpower_4ohm_22-14kHz_BW45kHz.png


There is not much difference from THD measurement, except for rising plots below 1W because of noise component in THD+N.

And the best for the end - comparison of THD+N vs. power at 1kHz measured at 1kHz with BW of 22kHz and 45kHz
UcD180_thdnpower_4ohm_1kHz_BW22-45kHz.png


There is no difference! Just a slightly higher noise measured at very low power due to higher measurement bandwidth.

Frankly - though the distortion is a bit high, I like this design for its consistent behaviour and no need for excuses on measurement bandwidth to be reduced to 20kHz.
 
More - let's increase measurement bandwidth of UcD180HG to BW = 80kHz and also let's eliminate the measuring low pass RC filter not to attenuate distortion spectrum components.

Measurement of 15kHz distortion

UcD180 BW80kHz 15k 75W4R.png


same in linear frequency scale
UcD180 BW80kHz 15k 75W4R lin.png

Please note the spectrum is clean, except for harmonic components!


And THD vs. frequency at 75W/4ohm
UcD180 BW80kHz thdfreq 75W4R2.png

Please note the dominating H3, 2rd harmonic component.

I am saying that HF linearity of the old UcD180HG is better than that of NCore and Purify.
 
We have been discussing in another thread the definition of measuring bandwidth for power amplifier distortion measurements, with conformance of practices used for decades and also with standards like IEC 60268-3.
https://webstore.iec.ch/preview/info_iec60268-3{ed5.0}b.pdf
The key is that distortion measurements with bandwidth of some 20kHz loose any technical reason and point above about 4kHz, because only few harmonic components are then taken into account which absolutely cannot reflect non-linearity of the amplifier. For decades, the common practice was to use at least 80kHz measuring BW. This has changed after the introduction of class D amplifiers, that were unable to compete under standard conditions of distortion measurements.

As a comparison, below please see THD+N vs. power measurements, into 4 ohm load, of Hypex NC252MP amplifier and of my A250W4R class AB amplifier (which has just average parameters for the class), measured at frequencies from 1kHz to 20kHz, with 90kHz measuring bandwidth. I think the plots speak for themselves. With the higher measuring BW, the class D amplifier cannot compete to an average class AB amplifier.

A250W4R_vs_NC252MP_THDNlevel_BW90kHz.png
 
Instead of analysing what is inaudible above 20 kHz, how about focusing on what is audible below 20 kHz?

IMD 18 + 19 kHz 10W into 8 ohms by SoundStage! measurements (Audio Precision APx555 B Series).

Bryston B135 Cubed
Bryston B135 cubed.png


Musical Fidelity M6si
Musical Fidelity M6SI.png


Rotel Michi X5
Rotel Michi X5.png


Rotel RA-1572MKII
Rotel RA1572 Mk II.png


Simaudio Moon 860A v2
Simaudio Moon 860A v2.png


Technics Grand Class SU-G700M2
Technics Grand Class SU G700 mk2.png


NAD M23: powered by Purifi Eigentakt, class D, and the cleanest of them all.
NAD M23.png
 
Instead of analysing what is inaudible above 20 kHz, how about focusing on what is audible below 20 kHz?
Because measuring what's audible will lead to conclusions that don't benefit advocacy of old-fashioned amplifiers.
 
NAD M23: powered by Purifi Eigentakt, class D, and the cleanest of them all.
NAD is an exception to the rule. Bruno Putzeys' Purifi Audio’s Eigentakt is not "just another class D". They call it "Hybrid Digital".
IMO their products are underated. Maybe their distribution and advertising model is not that great... IDK. But in US basically they don't have a presence, only in Canada.
 
Last edited:
NAD is an exception to the rule. Bruno Putzeys' Purifi Audio’s Eigentakt is not "just another class D". They call it "Hybrid Digital".
IMO their products are underated. Maybe their distribution and advertising model is not that great... IDK. But in US basically they don't have a presence, only in Canada.
Purifi Eigentakt is not "Hybrid Digital". It is an analog in, analog out class D module, period.

I think you are mixing it with NAD's "Hybrid Digital", which is their proprietary digital in, analog out class D amplifier, like the C 328 for example AFAIK.
 
Yes. All those advertiser-invented words confuse me...

I totally agree with their statement here:
  • Negligible Intermodulation Distortion (IMD)
    • A very good measure for how well an amplifier handles complex signals. Sonically low IMD means a highly resolved and stable stereo image across the whole spectrum, even during very complex and busy passages.
People keep saying that THD+N is all you need to know, because one sinus signal is sufficient.
I keep saying that that's false. Look at their DIM at >1W. Others don't even dare to publish that. Or they publish only the IMD CIF (10x lower).

PS: Same applies to delta-sigma DACs (similar principle of operation). People can't believe that they can sound marginal even if they measure great. But I digress....
 
Last edited:
People keep saying that THD+N is all you need to know, because one sinus signal is sufficient.
Who, specifically, has said that?
PS: Same applies to delta-sigma DACs (similar principle of operation). People can't believe that can sound bad even if they measure great. but I digress....
Who, specifically, has tested the claim of "sound bad but measure great"?
 
Who, specifically, has said that?

Who, specifically, has tested the claim of "sound bad but measure great"?
Yeah, funny how Amir's reviews, who does keep the SINAD chart, has all these other tests. You'd think his reviews consist of one test for SINAD at 1 khz and that is it.
 
Yeah, funny how Amir's reviews, who does keep the SINAD chart, has all these other tests. You'd think his reviews consist of one test for SINAD at 1 khz and that is it.
Show me where does Amir test IMD or DIM for amplifiers.

And regardless, I was talking about manufacturers of class D amps not showing those numbers.
 
Last edited:
Show me where does Amir test IMD or DIM for amplifiers.

/32 tones not good enough?
//He generally does that test
 
Show me where does Amir test IMD or DIM for amplifiers.
Note the multitone. That will show up any IMD. So your assertion is not only wrong, it's ludicrous.

And regardless, I was talking about manufacturers not showing those numbers.
Please name a manufacturer who has said "THD+N is all you need to know, because one sinus signal is sufficient" as you claimed.

Also still waiting on any specific example of a "sound bad but measure great." I suspect you can't come up with a single one, but I would be happy to be proved wrong.
 
Show me where does Amir test IMD or DIM for amplifiers.

And regardless, I was talking about manufacturers.
He did start measuring multitone after some time.

Examples:
index.php


index.php


index.php


index.php




But that is off topic, the point here is that high bandwidth to capture ultrasonics is as useless as evaluating sports cars on how fast they can drive on reverse lol.
 
Show me where does Amir test IMD or DIM for amplifiers.

And regardless, I was talking about manufacturers.
He tests one version of IMD (I prefer the twin tones at 18+19 khz), he doesn't test for DIM don't know I've seen anyone who regularly does. He does do a multitone which would show IMD even if not at the specific frequencies in the DIM test. He tests for FR, distortion at different levels and other things. The point being not that his tests are totally comprehensive, but that they are far more than a one shot one SINAD number and call it done. Yet you posted, "People keep saying that THD+N is all you need to know, because one sinus signal is sufficient." Like SIY, who says this? I only see people claiming other people saying this not anyone who does say this.
 
Once you've climbed the "I am right and everyone else is wrong" pole it is very hard to get back down.
 
Back
Top Bottom