The performance of different digital modes can be compared in numerous ways. The comparison most
expected by uninformed users is 'sensitivity', i.e. performance in the presence of white noise,
which can be measured with reasonable repeatability
using an ionospheric simulator.
However, this measurement is generally meaningless as so many
other factors come into play when considering HF performance - things like the susceptibility to
ionospheric Doppler effects, sensitivity to impulse or burst noise, and sensitivity to multi-path
reception, where there can be selective fading and timing differences between paths. However, these can
also be measured with a simulator, provided care is taken in choosing appropriate simulations.
Several other comparisons prove interesting and certainly less controversial, since the performance
of each mode can be calculated. The factors which are easily compared are bandwidth, bandwidth efficiency
and coding efficiency.
Rather than providing a series of measurements that would be difficult to understand, the information
shown below was condensed from a detailed study made of representative popular modes, using the
standard CCIR ionospheric simulation known as 'Mid Latitude Disturbed NVIS'. This is a near approximation
of poor multi-path conditions, without static or other interference, but with white noise reducing
the signal to 0dB S/N in 2.4kHz bandwidth.
It should be stressed that simulation results can be highly subjective, and are usually presented as
comparative text results for the observer to assess. A method has been evolved which involves counting
the number of correctly received complete words and the total number of words, and giving the ratio
the value '% Copy'. This technique has proved to be repeatable. From practical experience a QSO is
viable if '% Copy' is on average greater than 90%, and impossible below 70%.
From this graph it is clear that under these conditions (typical of 80m at night) PSK31 gives poor results,
and RTTY is not much better (remember this is for this specific simulation - for others the results will be different). The error corrected modes all perform well. Not how well DominoEX8 performs - with no
error correction! This mode was designed for these conditions.
The horizontal axis of this graph is typing speed, and you can see that MT63 is the clear winner.
However, the graph tells only part of the story - MT63 also consumes much more bandwidth than most
other modes. Olivia is the same bandwidth as MT63, but look at the low typing speed!
'Error Rate' as shown in this next graph is simply 100 - '% Copy'. Thus an Error Rate below 10%
will give comfortable copy during a QSO. The data has been rearranged to show (on the horizontal
axis) a measure of how effectively the mode manages its bandwidth. Clearly PSK31 uses the least
bandwidth for its typing speed, while Olivia is the least efficient, but at least has a low error rate.
Again, it must be stressed that the measurements are for 'Mid Latitude Disturbed NVIS' and the
modes which perform poorly may well perform better under other conditions. Anyone with two
computers and a copy of the AE4JY 'Pathsim' silulator can repeat these measurements for any
appropriate simulation conditions.
There is no controversy, subjectivity or choice of measurement conditions involved in comparison
of bandwidths of different modes - it all comes down to mathematics and modulation theory.
The 'Necessary Bandwidth' is defined by the CCIR using a set of mode-related formulae.
Clearly the narrowest bandwidth mode considered here is good old CW - Morse code, with PSK31
a close second. The widest modes are MT63 and
Olivia, both at 1kHz. However, this graph doesn't tell us everything, it is purely arranged in order of
bandwidth. Let's look at this bandwidth in relation to the typing speed of each mode.
The modes at the bottom left are the best performers, having a low bandwidth for their typing speeds.
MT63 doesn't look so bad in this comparison, as the typing speed is high (but who can type at 100 WPM?).
Olivia is the least bandwidth efficient by a wide margin. THROB achieves high efficiency because
it has a very small character set, while the MFSK varicoded modes also have good bandwidth efficiency
even (in the case of THOR and MFSK16) with FEC halving the typing speed.
The last graph gives an indication of how well each mode transfers data with each transmitted symbol.
This measurement is a combination of the size and efficiency of the alphabet used, the type of modulation,
and the rate of any FEC used. Clearly modes with low FEC rates have the highest redundancy, and so should
have the best error correction performance - but at the cost of data transfer speed.
The coding gain term used here is simply a measure of the typing speed compared with the symbol rate.
You can see that
the mode with the most redundancy is Feld-Hell, effectively 49 symbols per character. MFSK16 and THOR
occupy the middle ground, as they have FEC, while DominoEX, with no FEC has twice the coding gain of THOR.
The clear winners are THROB and MT63, but for completely different reasons - THROB has an extremely
limited character set, and can achieve one character per symbol (the only mode to do so), while MT63
has the advantage of 64 carriers, and so despite the FEC rate of 7/64, achieves a high coding efficiency,
of course at the cost of bandwidth.