On 8/17/11 4:24 PM, Alan Cresswell wrote:
That's interesting. I have collected all my passes on the TS2000 with the AGC on and set to the longest setting. This is mainly because I often record the signal level every 0.5 seconds during a pass which requires the AGC to be on and the longest setting irons out any short fades.
The main thing is that the gain remain constant, or nearly so, during each fade so that the random series of 0's and 1's produced during the fade on thermal noise is mostly seen as 'weak' 0's and 'weak' 1's. Increasing the gain during a fade makes those random bits seem stronger and more certain when they're still just random bits. That can make it harder for the Viterbi decoder to correct them as errors.
A Viterbi decoder works much like a network routing algorithm that looks for the cheapest path to a destination. It finds the best path through a 'trellis', a pattern of links corresponding to all possible state transitions in the convolutional encoder that produced the transmitted signal. Out of every possible path the decoder finds the one that most closely matches the received sequence and declares it as the one that was most likely sent. It can still be wrong, and when it does it usually emits a burst of several dozen errors until the decoder gets back on the right path. In BPSK-1000, this error burst causes the HDLC decoder to abort the current frame or discard it with a CRC error.
The Viterbi decoder tallies up the 'cost' of each link in a path to find its total 'path cost'. If a particular link assumes that a '0' was sent, then receiving a strong '0' is the best possible match so that results in the lowest possible cost for that link. A weak '0' gives a greater cost, a weak '1' an even greater cost, and a strong '1' gives the highest possible cost.
Even if one link in a path has a high cost, the complete path will still be chosen as the winner if all the other paths are worse. When this happens, typically all the other links in the path will closely match the received sequence.
So you really want to avoid classifying bits as 'strong' when you know the signal is gone and the bits can't possibly be right. That means not boosting the gain (and the noise level) during a fade. The noise level should be kept constant.
-Phil