Differential PSK should be more reliable in the presence of frequency drift and Doppler spread. There are two ways to do this: 1) compare the phase with the previous phase of the same subcarrier or 2) compare the phase with the phase of the next higher or lower subcarrier. In the first case, the first symbol transmitted is always all zeroes. In the second case, there would have to be at least one pilot subcarrier that is unmodulated. If you want pilot subcarriers, it should be possible to put them 62.5 Hz above and below the outermost data subcarriers as they take no extra space if they are not modulated.
A good way to do FEC is to use trellis-coded modulation (TCM). One bit is added to the data stream for each subcarrier. This, 1 data bit is sent using QPSK and 2 data bits are sent using 8PSK. The advantage of sending the data and ECC bits on one subcarrier is that the error-correcting code can be designed so that no extra bandwith is needed and that the addition of the extra bit actually decreases the required SNR rather than increasing it as you would first expect by increasing the number of points in the constellation. Ungerboek came up with a set of codes that can decrease the required SNR by 3-6 dB (with no fading) when going from QPSK with no ECC to 8PSK with ECC. The improvement is larger when fading occurs. The amount of improvement provided by TCM depends on the complexity of the state machine used to generate the ECC bit. However, a simple algorithm with 4 states provides a 3 dB improvement. A Viterbi decoder is used to calculate the most probable set of state transitions that the incoming signal has taken from symbol to symbol and then backtracks to determine the most likely combination over an entire data frame. It can also make decisions based on the actual value of the incoming signal rather than on 3 already decoded buts. This adds another 2 dB of improvement. Its probably useful to place the audio subcarrier frequencies in the 500-1000 Hz range or higher so that harmonics of low frequency subcarriers don't interfere with higher-frequency subcarriers. 73, John KD6OZH ----- Original Message ----- From: Rud Merriam To: digitalradio@yahoogroups.com Sent: Wednesday, October 24, 2007 10:03 UTC Subject: [digitalradio] OFDM Proposal: Details For your amusement and consternation here are my latest thoughts on doing an OFDM protocol. Symbol rate: 62.5 Hz (128 samples @ 8000 Hz) Guard interval: 2, 4, 8 ms adaptive to conditions Subchannels: 8 (62.5 125 187.5 250 312.5 375 437.5 500) Bandwidth: 437.5 Hz Raw BPS: 1778, 1600, 1333 adaptive (guard band change) Base frequency: undetermined MODULATION (somewhat firm) Waveform: DQPSK with constellation at 45, 135, 225, 315 degrees Generation: 8 separate generators providing continuous waves through the guard bands Phase change: start of symbol period Shaping: post generation raised cosine over symbol and guard period DEMODULATION (somewhat speculative) FT: 128 bin every 32 samples for locating subchannels Synchronization: square of subchannels identified by FT to locate bottom subchannel by 125 Hz signal Frequency drift: subchannel selection based on output of synchronization Phase detection: phase averaged over symbol period, differential with last symbol A main goal is to keep the bandwidth within 500 Hz. The symbol rate is as suggested by John KD6OZH. First testing will probably be with his 8 ms guard band but I would like to make it adaptive to short that period if multipath conditions allow. DQPSK to get more throughput and because getting the absolute phase is a challenge. Any suggestion to use absolute phase would be appreciated since that gains a couple dB. The Fourier transform is mainly to identify the potential subchannel locations to allow adjusting for frequency drift. Once high energy bins are determined the signal is filtered at various of those frequencies and the square used to detect the doubled lowest frequency (125 Hz). That also locates the symbol period for synchronization. Actually, the possible frequency includes the guard band so it may be one of three values. By determining that value the guard band period is also determined and the actual guard band removed. Rud Merriam K5RUD ARES AEC Montgomery County, TX http://TheHamNetwork.net