The DSP software radio package by SM5BSZ can be configured for a wide range of compromises between performance and computer load. The current test release is just a very early test for the first processing step, the first FFT. RADIO INTERFACE: fs=sampling frequency B=system bandwidth The radio can be interfaced in 3 different ways): 1: One channel audio. This means you have a just an ordinary radio and connect it's audio output directly to the computer. Max bandwidth B can approach 0.5*fs if the filter of the radio is very steep so frequencies above 0.5*fs+B are sufficiently suppressed. Normal radios have good filters so this mode can be used with slow computers even without audio filtering to remove signals above 0.5*fs 2: One channel I/Q. The audio board is run in stereo and the two channels should be connected to the output of two mixers that are fed with local oscillators having a phase shift of about 90 degrees. Max bandwidth is can approach fs if very steep audio filters are used between the mixers and the computer. 3: Two channels. This means you have a radio with two receive channels using common oscillators. (stereo receiver) Max bandwidth can approach 0.5*fs with very good filters. In this mode the two channels are automatically combined for optimum S/N which is very useful for e.g. EME where the two channels normally are the two signals from a X-yagi. The process completely eliminates signal degradation due to farady rotation. Linear elliptic and circular polarisations are automatically received with the optimum receiver polarisation. On HF bands the two signals can be any two equal antennas. The adaptive combination of the two signals for optimum S/N is can then be adaptive direction finding or adaptive polarisation depending on the antemnna geometry. Noise suppression is greatly enhanced in this mode since interference sources are evaluated with the combination of the two signals that maximise the S/N for each particular interference source. The contribution to the interference level in the signal channel can usually be removed with better accuracy compared to single channel modes. 4: Two channels I/Q. A combination of modes 2 and 3. This mode is for high end audio boards using 24 bit A/D converters with 4 channels. (not yet implemented) SAMPLING SPEED: The sampling speed must be set high enough to avoid alias signals. If you can tolerate alias signals at -60dB you can place the sampling frequency halfway between the -10dB point and the -70dB point of the low pass filter you have between the radio and the PC. Remember there is usually a low pass filter inside the audio board and another one in the receiver itself so you do not have to add fancy filters unless you want the very large bandwidths you can get by pushing e.g. the -10dB and the -70dB close together. If you can afford the processing time you can improve dynamic range by oversampling. Set the sampling speed as high as possible with the computer you have. SPECTRUM RESOLUTION AND CPU TIME: The times for the CPU intensive routines are sensitive to processor type, cash memory size and of course clock frequency for processor and main memory. If you select a narrow bandwidth you get a large size of the FFT. When the FFT is too large to fit in the cash memory there is a severe punishment in processing time. If you really want the frequency resolution you may set the sampling speed for no oversampling i.e. as low as possible before you get aliasing signals. There is no such thing as the optimum code for FFT routines. The PC radio is designed to accomodate a selection of different FFT routines There is a function in the set up that will allow you to select the fastest one on your hardware for the FFT size you have selected. The FFT time is sensitive to the selected window function. A narrower window makes the FFT size larger for any given bandwidth. At the same time the transforms overlap more in time so the number of transforms per time increases for a given FFT size. If you want to track weak signals in white noise you may select a very wide window sin power = 0 means no window and no overlapping of transforms. If you have strong signals within the passband you may get them to fall off much more rapidly with a narrow window even if you have to set the bandwidth higher to avoid excessive CPU load. WIDE BAND PROCESSING: The first FFT produces a spectrum on the screen. You may use that directly to locate and lock to interesting signals. If you use a normal SSB receiver with a builtin noise blanker that operates in a larger bandwidth you probably get little improvement by enabling the second FFT. If the second FFT is enabled, the transforms of the first FFT are split into two sets of transforms. Both sets are back transformed to produce two functions of time, one of which contains all strong signals (anything that looks like a peak in the spectrum) while the other contains impulse noise and weak signals. The strong signals are amplitude limited by a selective AGC function to make sure 16 bit will be enough for further processing. The back transform of the first FFT that is associated with enabling of the second FFT allows a very efficient noise blanker, particularly in stereo at large bandwidths. The second FFT that uses the sum of the two time functions from the first FFT can be set for very narrow bandwidths because a frequency selective AGC function makes it possible to use MMX instructions to improve speed. AVERAGING AND AFC To find and follow weak signals in noise averages of transforms are used to improve S/N. For very weak and unstable signals it is possible to use an extra delay so the signal can be found by use of the spectrum both before and after the processed moment of time. In case second FFT is enabled, AFC uses second FFT transforms. Otherwise first FFT transformas are used. AFC averaging time is limited by the time allowed for the transforms used. FIRST MIXER The selected signal(s) are frequency mixed with a local (digital) oscillator that follows the frequency determined by the AFC algorithm. In this way unstable signals get a reduced bandwidth centered around zero frequency. Rather than actually mixing the original time function with a digitally controlled oscillator (NCO, looking up sin ans cos values in tables ) and then filtering the mixer output (I and Q, complex signal) to allow resampling at a lower sampling rate, limited back fft's are used. Exactly as in a conventional mixing and data decimation process spur suppression is important. The filter in use during this process is the window function of the FFT. To get good spur suppression the window power of sin has to be at least 2. The spurs depend on the first mixer data reduction rate and the size of the transform. FURTHER PROCESSING FOR WEAK CW Once the bandwidth is reduced processing is less time critical. For weak CW a second AFC plus second mixer takes advantage of the reduced bandwidth produced already. This way the signal can be followed over weak periods where the first AFC only was succesful on surrounding peak amplitude regions causing a bandwidth reduction by interpolating the frequency. Finally the phase and amplitude is extracted from a very narrow filter and used for coherent cw. This is the processing sucessfully used in the DSP program for MS-DOS. The time delay can be selected from a few seconds up to about 10 seconds in the most difficult cases. FURTHER PROCESSING FOR NORMAL CW A normal signal can be followed by looking at the history so processing delay is limited to the delay caused by the filter bandwidths an a small extra delay for some safety margin in buffers. FURTHER PROCESSING FOR METEOR SCATTER Similar to weak signal CW but much larger bandwidths and correspondingly shorter delays. Coherent CW and automatic translation to ASCII are obvious additions that will work well for this propagation mode. FURTHER PROCESSING FOR FM Very large enhancements are possible compared to a conventional FM receiver. Maybe not so exciting because when FM is not good enough a change to SSB will be a good alternative................. FURTHER PROCESSING FOR SSB Decoding what happens in neighbouring channels to calculate the phase and amplitude of splatter from undesired signals so it can be removed should allow reception of SSB under very severe interference conditions. This is completely untested. For other modes I have some experience.....