The key to the digital signal processing lies in the FFT (Fast Fourier Transform). See Figure 10. The FFT can be seen as a large number of bandpass filters. With this, we can divide the audio band from 0 Hz to 4 kHz in 256 bands of 15.625 Hz. Because the bandpass filters overlap each other somewhat (spectral leakage), the net frequency resolution becomes about 30 Hz. If we increase the FFT frequency resolution, this will be associated with a longer time window. The frequency content of the sources varies in time and the chance of a contribution from an interfering source into one of these bands will remain roughly the same. But the average amplitude of this contribution decreases accordingly. So, a higher frequency resolution gives progressively better results. A four times higher resolution (7.5 Hz) is feasible at the cost of an increased time delay in the processing.
Figure 10 — Block diagram of the digital signal processing.
Apart from amplitude, the FFT also calculates the phase of each frequency component and it is therefore simple to calculate the phase difference. In the “windows” block, the software determines if the phase difference is within a window, and is calculated, depending on the form and function of a window, how strong the respective component must be passed or suppressed. The “select” block will calculate the final strength of the frequency components. The “IFFT” (inverse FFT) block converts it into a signal in the time domain.
With the “equalize” block, at the beginning of the block diagram, we can correct for differences in both receivers using an adaptive filter. With the noise source connected to both receivers at the same time, an adaptive filter adjusts the frequency characteristic in such a way that the sub receiver signal equals the signal from the main receiver.
The CW-filter has a somewhat hidden function. We decide per unit of time whether a frequency component may be passed. A passed frequency component during this unit of time has a constant amplitude output and is hard switched on and off. This produces the same effect (key-clicks) as that of a hard switched CW signal. With this CW filter, we reduce these “clicks.” With the bandwidth used for SSB, this phenomenon is not audible.
Last update: September 24, 2006