I was doing some loopback tests to get a grip with the analyser. Testing the AMP frequency response at about 96Khz I get the following plot. My quesiton is is the loss above 20kHz a feature of the analyser or is it my cable? I am using a 50 cm long BNC/BNC RG-58/U cable.
I took the QA403 on my desk and ran 3 sweeps, starting with 64K FFT for 48Ksps sample rate. And 96K I used a 128K FFT, and at 192K I used a 256K. For the sweeps, you can see the 48K has a bit of peak (0.01 dB at 20 kHz) and then drops off sharply. That is purely the digital filter inside the ADC.
For the 96 and 192K sample rates, you can see a bit of what looks like a first order filter rolloff before the sharp drop. That is an RC in the front-end of the analyzer. At 96K, you should be down about 0.1 dB at 37 kHz. And at 192k, the 0.1 dB point comes a bit sooner, but the 0.5 dB point comes a bit later.
Hi @DavidR. I think the error is because you also need to set at least 40Khz in the measurement options window, which is probably now set to less than 40Khz
Yes, Claudio is correct. The error code hopefully is a reminder. Note the Measurement Stop Frequency is how you specify where the measurement will stop looking for harmonics. If you set the Generator Stop Frequency to 40k and your Measurement Stop Frequency to 50 kHz, so you won’t see any harmonics included in the measurement.
And so, if your aim is to measure the second harmonic of 40 kHz, then you probably want your Measurement Stop Frequency at 85 kHz, sample rate at 192k, and your Generator Stop Frequency to be 40 kHz.
Maybe due to the phase change caused by the eq? Exponential chirp methods assume equal phase delay for every frequency as they correlate the chirp in against chirp out under the assumption that frequency is a fixed function of time. A longer chirp will decrease the effect of this if so.
Hi @cfortner, yes @MarkT is on to it. An exponential chirp spends as much time sweeping from 20 to 40 Hz (one octave) as it does 10k to 20k (also one octave). So, if we are sweeping at, say 1 octave per second, then the dwell time on each 1 Hz slice in the first octave is 1/20 = 0.05. And the dwell time on each 1 Hz slice in the last octave is 1/10k = 100u.
So, there is 0.05/100u = 500 times more energy in the first octave than the last octave. And so, you are seeing the uncertainty that comes from the much lower energy. You are at 192k, but you are only looking at the first 20 kHz. If you cut your sample rate to 48k, you’ll spend 4X more time at each frequency which should help with noise. And then increasing FFT will help too.
And unless you expect a big difference in EQ at mild versus hot levels of output, you might skip the attenuator and adjust the amp to give about 0 dBV of peak output. And then use 6 or 12 dBV full scale. And once you are satisfied with the shape of the curve at the lower level, crank it up and see how it changes, knowing that the added atten will increase the noise AND uncertainty.
Notice that in the first graph the dbV range is 120dB, while in the second graph it’s 35dB: I guess that the high frequency noise would be visible in the first graph too, if you put a 35dB range.