I almost always use the Chirp frequency response method, which seems to give me good results on the gear that I measure- amps, receivers, preamps and integrated amps. Occasionally I have gotten some results- usually on tube gear with output transformers- where I had to play with bandwidth, sample size and the FFT size to get an “optimum” frequency response. I was testing an “all in one” class D amp over the weekend and got this frequency response using the Chirp method (changing sample size, FFT and bandwidth did not affect the response much):
I manually checked a few points at the very low end of the band and they agreed with the AmpFrequencyResponse test. I was wondering what method was preferred and why (aside from the Chirp method being faster)…Thanks for your input
Hi @Var, chirps are great for a quick look. But remember most all systems will have AC coupling caps, and chirp can be a fairly abrupt start that might cause some weird offsets to propagate through in a DUT. So, some DUTs will cope fine, others will not.
Can you post the output and input time domain of the chirp when you are seeing the ringing show in your first plot?
Thanks @cfortner- I have found the chirp method to work rather well, but unfortunately, did not try it with the AmpFrequency Response method when the results were not great. @matt- I have something else on the bench now but will retest the unit and get the two time domain plots. I do not have a schematic for the device so can’t about input caps, but that is a point I did not consider. Thinking about the AmpFreqResponse test, could it have an option to show the response as dBR instead of gain ?
Is the time domain output plot the QA403 output direct, or the QA403 output when the all in one class D is connected and powered up? The reason I ask is the input impedance vs frequency of the amplifier can cause some of the deviations (not saying that is true of this amp).
With a lot of these modern low input impedance amplifiers floating around, I’ve taken to splitting (T-BNC) the output of the QA403 to send to the unused channel as a sanity check if you’re just testing one channel at a time.
I’m wondering how much of the LF issue is caused by DC offset. I had a hell of a time with my old analyzer due to DC throwing FR plots off. The coupling cap to the ADC had to be “understood”…
The thing with Chirp is that it runs a single test with a signal that covers your entire frequency range. As you can imagine, that doesn’t allow any one frequency very much time in the test. So, for example, 10Hz in your test is getting maybe 1/20th of a second or 1/2 a cycle if you’re lucky. This causes accuracy issues at the low end (and high end).
If you want to use chirp but improve the accuracy it helps to use a longer time period (increase the fft size) and maybe reduce the number of octaves the test spans.
@MarkZ- thanks for the input- I tried several different FFT’s, up to at least 256K w/out the response changing much. Since I am not in any time constraints with what I do, I don’t mind running the automated test- I just never gave it much thought with the chirp method working well (most of the time). I never have done anything with the # of octaves. For grins, here is a recent plot from a Cary tube preamp from the 90’s:
@restorer-john- The class D amp, I am a little embarrassed to say, is a Fosi MC331, which has a pair of tubes in the preamp stage, and bluetooth etc integrated amp. It was powered on for probably 20minutes before any testing. Unless there is an output power spec where the power is specified for just 1 channel driven, I typically test both channels at the same time.I am not sure I follow what you are describing with the unless it is like using the Right channel as a reference. I suppose I could look at the input impedance with the QA462 at some point.
Ignoring everything else… that’s a tube preamp? Pretty impressive. I don’t want to get off-track though.
I think also that the circuit, as @restorer-john is alluding to, has a pretty big input capacitor and when you first start pumping voltage into it there will be bounce for a short time period - enough to influence those low-frequency measurements.
