MT63 is intended for
conversational use between one of more Amateur Radio stations, providing good
performance under poor conditions, and consequently utilizes FEC rather than
ARQ error correction processes.
The MT63 modem, constructed around a high speed DSP processor, either in a
dedicated external DSP unit like the Motorola EVM, or in PC software using
the PC sound card, transmits 64 tones spaced 15.625 Hz apart, in the 1 kHz
bandwidth. The base-band signal occupies from 500 Hz to 1500 Hz. All 64
tones are differential bipolar phase shift keyed at 10 baud. Since the Walsh
FEC code is 64 bit, the character rate is the same as the symbol rate, so the
throughput with FEC is ten 7-bit ASCII characters/sec (about 100 WPM). The
following diagram (courtesy of Eduardo EA2BAJ) illustrates the 1 kHz wide
MT63 Power Spectrum
MT63 is intended for conversational use between one of more Amateur Radio stations, providing good performance under poor conditions, and consequently utilizes FEC rather than ARQ error correction processes.
The MT63 modem, constructed around a high speed DSP processor, either in a dedicated external DSP unit like the Motorola EVM, or in PC software using the PC sound card, transmits 64 tones spaced 15.625 Hz apart, in the 1 kHz bandwidth. The base-band signal occupies from 500 Hz to 1500 Hz. All 64 tones are differential bipolar phase shift keyed at 10 baud. Since the Walsh FEC code is 64 bit, the character rate is the same as the symbol rate, so the throughput with FEC is ten 7-bit ASCII characters/sec (about 100 WPM). The following diagram (courtesy of Eduardo EA2BAJ) illustrates the 1 kHz wide MT63 spectrum:
There are two other bandwidths that can be used, 500 Hz, and 2 kHz, where the tone spacing and baud rate are halved or doubled, and the throughput halves or doubles respectively. Unless otherwise indicated, this description is of the default 1 kHz version.
In addition, an optional doubling of the interleave period improves the temporal resistance (e.g. to burst noise) at the expense of increased time delay through the encoder and decoders. The different speeds are achieved by scaling all the timing factors, although the lowest carrier frequency remains constant at 500 Hz.
|Bandwidth||Audio Range||Symbol Rate||Char Rate||Interleave/char|
|500 Hz||500 - 1000 Hz||5 baud||5 char/sec||6.4 or 12.8 sec|
|1000 Hz||500 - 1500 Hz||10 baud||10 char/sec||3.2 or 6.4 sec|
|2000 Hz||500 - 2500 Hz||20 baud||20 char/sec||1.6 or 3.2 sec|
The user data from keyboard or file, (the data code is 7-bit ASCII), is further encoded into 64 bits using a Walsh/Hadamard function to provide a highly robust FEC technique with high redundancy. The Walsh function ensures that up to 16 of the 64 bits can be corrupted, yet decoding will still produce an unambiguous result.
The MT63 signal is spread both in the time domain (temporally) and the frequency domain (spectrally). To ensure that noise bursts and other time domain interference artifacts have minimal effect, each encoded character is spread over 32 sequential symbols (3.2 sec). To ensure that frequency domain effects, such as selective fading and carrier interference have minimal effect, the character is also spread spectrally by using all the tones across the width of the transmission.
In a "long interleave" option, the spreading is over 64 symbols (6.4 sec), with consequent improvement in resistance to impulse and periodic interference, but of course double the time taken for the data to "trickle through" the Walsh encoder and decoder pipeline.
The above picture, which was captured using waterfall plot software GRAM, shows about 10 seconds of a 10W MT63 transmission on 14 MHz over a 3000km path, clearly showing bands of selective fading passing through it, but with no noticeable effect on copy. (Horizontal axis is time, vertical axis is frequency). The signal had a slowly falling wooshing sound characteristic. The selective fades can be faster or slower, up or down, but do not adversely affect copy. This pattern is typical of single-hop DX or local station NVIS conditions.
The example below is from the same station on the same evening, but on 80m, with much more confusing and complex fading. The exact 1kHz bandwidth is still obvious. Copy was still 100% under these conditions, although more power (50W) was required.
No special tuning technique is required because the signal capture logic is capable of locking with ±50 Hz frequency error (±80 Hz in the latest software). The confidence of the FEC correction system is degraded as mistuning is increased, beyond this limit, just as it would be if some of the 64 carriers were obliterated. The tracking logic will track timing and frequency error indefinitely. The decoder uses FFT techniques to define "buckets" into which carrier phase information is collected. It uses differential carrier phase detection to track phase changes introduced as a result of ionospheric variations.
With 10 Hz baud rate, the effect of ionospheric doppler on signal phase within one bit time can be very serious (even more than PSK-31 because of the longer bit time), and as a result, MT63 is not a particularly good performer in this respect. The performance under these conditions is still subjectively better than PSK-31 however, because of the huge redundancy provided by the FEC system.
The receiver utilizes 64 parallel channel differential bipolar phase detectors, which ignore both frequency and amplitude variations. These provide soft solutions to the 64 Walsh decoders. Since all 64 channels generate Walsh function solutions at the same time, i.e. there are 64 parallel receiver demodulators and Walsh decoders, the receiving routine simply chooses the solution that gives the least error. This technique allows the receiver to avoid frequency ambiguity which will result if some tones are absent due to interference or fading, or if because of mis-tuning the tones are decoded as their neighbours. Newer software use further parallel receivers, allowing a greater range before mistuning degrades performance.
The receiver demodulators supply the decoders with soft phase solutions in the form of cos(phase_difference), to further enhance the accuracy of the Walsh FEC decoder.
There are of course many ways to encode the data across 64 tones and in 32 symbols. Pawel has tested many (but not all!) of the possible methods and chosen one that works well. There is much room for experimentation, and Pawel has made the source code public domain for this purpose.
MT63 code is available for EVM, Linux and several versions for WINDOWS™ 95/98/NT/2000/XP.