Class-D Amplifier Frequency Response Variability with Different Loads
The Pitfall of Cheap Class-D Amplifiers
Many Class-D amplifiers offer great specs and power output at affordable prices and come in very compact form factors. On paper and even on the bench using static loads they appear to be great performers however due to poor output / low pass filter optimization or inadequate feedback on the output many of these amplifiers can suffer from noticeable frequency response variability in the top octaves in the real world when driving common passive speaker impedance loads. Unfortunately this issue often goes unnoticed if the amplifiers are only bench tested using 4 or 8 ohm resistive loads which will not show the severity of the issue.
What this means is that if you are using a Class-D amplifier with a poorly designed output filter then the frequency response from the amp can change quite dramatically depending on the speaker you have hooked up. One speaker may cause the treble to be several dB too high another several dB too low. Thus you can't trust the amplifier will correctly reproduce the input signal. This frequency response swing often extends outside of the audio range as well and in extreme cases can be so severe as to cause amplifier instability
effecting operation and causing potential damage to the amplifier.
While not prevalent in all Class-D amplifiers I see very little testing for this issue so I do not know the extent to which this issue plagues these amplifiers, which are impacted and which are not. It certainly appears those on the budget side are more likely to be affected, though I would not rule out more expensive amps as a whole as the design of the output filter one some of those may be just as careless as on the cheap amps.
Basic block diagram of a class-d amplifier. The low-pass filter is there to filter out and smooth the very high frequency switching on which the amplifier uses to modulate the signal. Poor design of the filter can result in interaction of the filter with the speaker impedance affecting the frequency response within the audio range.
I started noticing these issues in listening well before I thought to test for them.
Several years back I wanted to listen to a set of speakers currently I had been working on in a different room, the nice compact Dayton DTA-100A I had just bought seemed like an excellent choice for a quick temporary setup. Only after hooking everything up I noticed right away something sounded off, the speakers lacked the smoothness I was used to and now seemed overly bright and grating. I knew the change in location couldn't cause that large an issue so I swapped over to my older APA150 (AB amp) and like I expected the irritating treble was gone and the speakers sounded great again. I figured the DTA-100A just didn't sound all that great (which was partially true) but I just stuck it on a shelf didn't dig too deep into why it sounded bad.
When I first really became aware of the exact issue was after I had purchased an AIYMA A07 following the generally positive review on ASR. It looked like a very decent amp for the price with great output power and was very compact, perfect for a nice small bench amp or so I thought. When I decided to try it out I wanted to compare noise and distortion performance to the Behringer A500 that I normally use. I hooked it up to the speaker I was working on at the time and noticed it certainly produced less audible noise from the sensitive compression driver in the speaker. Unfortunately when I ran a sweep in REW I thought what happened to my crossover, there was now a big upward swing to the treble. After several minutes double checking the crossover connections and trying to replicate the issue in the crossover modeling software I had the light bulb in my head go off, obviously it had to be the amp! I swapped wires back to the A500 and the frequency response was back to normal. I took a quick measurement of the output from the amps to compare and when the A07 was connected to a resistive load it was fairly flat, but when connected to the speaker the response swung upward in the treble by a sizable amount. The Class AB Behringer A500 on the other hand measured very similar with both the resistive load and the actual speaker load.
Testing for this issue:
I now decided to finally take the time and gather some more thorough and presentable measurements in order to illustrate how severe this issue can be. For these tests I'm driving the different loads with several amplifiers and measuring the output of each amplifier at the load itself using a symmetrical voltage divider connected to my 2i2.
A loop-back calibration from the DAC output to the 2i2 was performed to eliminate the small amount of rolloff in the signal chain (before the amp) when measuring up to the 40kHz range I used for these tests.
I noticed there is some small variability in the frequency likely due in part to losses from the speaker cable and connectors themselves that occurs on all the amps tested so be aware of that when looking at the measurements. Unfortunately I had to adjust the gain structure multiple times due to input clipping when I tested the class D amps so the responses have been manually aligned at 1kHz.
In addition to a standard 4 and 8 ohm resistive loads I am testing the amplifiers when connected to several speaker designs, the AlTi, Argos, Helix, Slipsteam and VBS-6.2. The impedance curves of these speaker designs are posted below for reference.
Amplifiers tested so far:
Aiyma A07 (Class D)
Here is where these poorly designed Class-D amps fall apart, this is not what I would consider an acceptable variation in frequency response from an amp with regard to the load/speaker connected.
Depending on the load there is a 4dB spread in output from the amplifier at 20kHz and it swings even more wildly beyond that range.
Audio Source AMP 100 (Class AB)
Very consistent, a max deviation of 0.4dB between the different loads which may be losses from the cable/connectors. Amp does seem to have around 1dB of roll-off by 20kHz.
Being Class-AB you do not need to worry about wild swings in the treble output depending on the speaker connected.
Behringer A500 (Class AB)
Note quite as good as the Amp 100, a little more variability at the bottom and top end but no more then 0.8dB at the extremes within 20hz-20khz and only +0.1dB/-0.3dB spread within 100Hz-10Khz again likely due in part to losses in the cable/connectors.
Dayton DTA-100a (Class D)
The worst of the bunch, an even larger spread then the A07 in the audible range, nearly 6dB of overall deviation at 20kHz and more then 2dB at 10kHz depending on the load.