How do you define linearity of amplifier? In simpler less-technical way!
Ok, I'll have a go!
The job of an amplifier is to make an electrical signal bigger. Ideally, the output signal should have exactly the same shape as the input signal, but with the voltage at each instant multiplied by the same scale factor (the gain). If the shape remains exactly the same then the amplifier is linear.
In reality the shape can get a bit distorted (i.e. the peaks might get slightly clipped, in which case they have been amplified by a lower gain factor). This is nonlinear behaviour because the amount of gain (the multiplication factor) depends on how large the input signal is.
When we calculate an FFT of a time domain signal, it is decomposed into its frequency domain components. If we added together sine waves in the proportions shown on the FFT (and with the correct phase) then we could construct a signal with the same shape in the time domain.
So for a perfectly linear amplifier with a pure 1kHz input signal the FFT of the output would only show a single spike at 1kHz. If the output is slightly distorted then this shows up as contributions at other frequencies in the FFT. When the input signal is a pure tone, the strongest distortion components typically occur at harmonics of the input signal frequency (i.e. 2kHz, 3kHz etc if the input is at 1kHz).
The other factor discussed is the bandwidth of the measurement. I.e. Which range of frequencies is considered when calculating the SINAD. I think Amir's default for this is 45kHz. Consider an amplifier that has a large distortion component at the 3rd harmonic.
If the test is run with a 1kHz input signal then the distortion at 3kHz is easily within the measurement bandwidth of 45kHz. The SINAD value will be degraded.
But if the input signal is at 20kHz then the 3rd harmonic at 60kHz is outside the measurement bandwidth and will not be taken into account in the SINAD value.
What’s your interpretation?
A highly simplified explanation of a Class D amplifier is that its output voltage is controlled by a feedback loop and two switches. One connects to the high voltage rail (increasing the output voltage) and the other the low voltage rail (decreasing it).
Class D amplifiers (the Buckeye NC252MP and the BoXem Purifi 1ET7040SA) have high internal switching frequency, but given the mode of operation, perhaps it wouldn't be too surprising if their amount of distortion increased as a function of input signal frequency.
The differences between the Purifi 1ET7040SA (class D) and Soncoz SGP1 (class AB) 15kHz SINAD results are perhaps due to differing spread of distortion components. Eg. The way that the Purifi module distortion seems to reduce at higher power levels is perhaps because its distortion tends to shift to higher harmonics and therefore gets pushed outside of the measurement bandwidth.
I don't think distortion components that are >20kHz are going to be an audible concern which is why I was wondering if limiting the measurement bandwidth to 20kHz might make sense.