Data comms well below the Critical Frequency

By: Ray Scrivens (G3LNM) and Bernard Spencer (G3SMW)

Background

There is, understandably, a lot of interest in using the 5MHz band for NVIS daytime emergency communications around the UK. However, the critical frequency does sometimes fall below 5MHz and it then becomes necessary to switch to the 3.5MHz band. The question therefore arises: why not use 3.5MHz all the time? The simple answer is that path losses are much higher at frequencies considerably below the critical frequency. This means that on days when the critical frequency is relatively high (say, 7MHz), the losses on 3.5MHz are likely to be too high for voice communication between typical amateur stations.

However, many digital modes can operate with signal levels some 20dB or more below that required for SSB voice. It is, therefore, worth investigating whether at these levels it may be possible to use digital modes in the 3.5MHz band almost consistently during daytime for inter-UK communications. To this end, tests have been carried out between Bernard (G3SMW) in Marlow, Bucks, and Ray (G3LNM) in Norfolk, approximately 15km east of Norwich. The path length is 196km. The tests were carried out on several consecutive days during February and March 2004, mostly at times between 12:00 and 13:00 UTC on a frequency of 3.585MHz. The digital modes used for most of the tests were PSK31, MFSK16, Throbx and Domino. The antenna used at G3SMW (the transmitting end) was a 21m horizontal doublet, 8m above ground, the wire running approximately North-South. At G3LNM the receiving antenna was normally a trap vertical although a horizontal doublet was tried on occasions. In terms of signal to noise ratio there was little difference between the two receive antennas.

Test Series 1

These tests were aimed at comparing the digital modes (i.e. PSK31, MFSK16, Throbx and Domino) to see which was the most suitable. The object was to assess communications reliability for a one-way path, such as would be used for emergency broadcasts. Separate test transmissions of about 2 minutes duration were made on each mode. The number of characters received in error were counted and compared for the various modes.

It was found that variations in signal strength due to fading and noise level changes from one test transmission to the next made it difficult to produce any consistent results. The "best" mode on one day would fare much worse compared with the other modes on another day. What these tests did show is that, in practice, the variations in signal level due to fading are much greater than the variations in sensitivity between the different modes so that the theoretically more sensitive modes (eg Throbx and MFSK16) seemed on average to produce only marginally better copy.

Test Series 2

In an attempt to get more meaningful results, an idea developed by Lionel G3PPT called “Combo” was tried. To create the Combo signal, firstly a digital audio recording of a test message generated by one of the standard digi-mode programs (e.g. Multipsk) was made for each mode (i.e. PSK31, MFSK16, Throbx and Domino). Care was taken to ensure that the peak amplitudes of the four recordings were the same. Different audio frequencies were used for each mode. The audio streams were then added together and recorded to produce the Combo multiplex. This combined audio signal was then played into the audio input of the SSB transmitter at G3SMW. To a listener, the Combo signal sounds like four separate simultaneous transmissions. At the receiver, a digital audio recording was made of the receiver output which was then subsequently analysed by replaying it, as required, into the decoding programs for the various modes.

The principal advantage of the Combo technique is that all signals are transmitted at the same time and are therefore subject to the same propagation effects. A difficulty with the technique is that, to maintain linearity in the transmitter, the peak envelope of the Combo multiplex must not overdrive the transmitter. This means that the peak power for each mode was only one sixteenth of the transmitted peak power. So, with a transmitted peak power at G3SMW of 50 watts, each individual signal had a peak power of only just over 3 watts.

Even with the Combo tests, fading caused problems in comparing the performance of the various modes. This was because the fading was frequency selective, so that when one mode was performing well, at the same time another could be in the trough of a fade (anyone who has watched the waterfall display of an MT63 signal will have noticed the moving diagonal stripes caused by selective fading). Again, it was apparent that fading had a much more significant effect on readability than the differences between individual modes. Even signals which were, on average, quite strong would occasionally dip into a deep fade. This may not matter for normal amateur comms, but for emergency comms it may be very significant. Users have become used to 100% reliable copy in e-mails etc.

Test 3

As a final test, a Combo signal was prepared with three PSK31 signals at 700Hz, 1kHz and 1.5kHz. The text messages on the three signals were identical and they started at exactly the same instant. Therefore, given good conditions, three simultaneous copies of the same message should be received. If selective fading is present, it is unlikely that all three signals would be in a trough at the same time, so that it ought to be possible to reconstruct the original message by picking the “good bits” from the individual streams. A test carried out on 15th March 2004 verified this although, because of the variable length characters in PSK31, it was difficult to cross correlate between the individual received message streams.

Findings

Fading was always present and all modes showed drop-outs over a transmission period of 2 – 3 minutes. Even when the average signal strength was high, the drop-outs still occurred, although of shorter duration (as is to be expected). The period of fading varied from 2-4 seconds on some days to in excess of 1 minute on others. The fading observed was always frequency selective with a spacing between nulls of around 1kHz (suggesting multipath with a path length difference of about 300km).
The received average signal strength varied from day to day. On the worst days, the signal was barely audible (at a transmitted power level of 3 watts) and copy was poor on all modes. There was no clear correlation between received signal strength and the “real-time” ionospheric data available. Our estimate is that using a 100 watt transmitter and “normal” antennas, useful communication would have been possible on 90% of the days.
The variations in signal level due to fading were much greater than the differences in sensitivity of the various modes. For one-way broadcast messages the crucial requirement is to have a good strategy to overcome this fading.
Although easy to tune, the Domino mode was nearly always the worst performer in terms of readability.

Conclusions

The tests suggest that a reasonably well equipped amateur station with a transmitter power of 100watts or more should be able to communicate on most days within the UK on 3.5MHz for either operational traffic or broadcast purposes using one of the existing narrow band digital modes (e.g. PSK31, MFSK16 or Throbx). Fading was found to be always present and some form of strategy is necessary to combat this. For two-way communications this can be overcome by simply asking for repeats or sending the message more than once when conditions are poor.

None of the existing digital modes are really suitable for very low error-rate one-way transmission when signal levels are fairly low and fading is present. ARQ techniques (e.g. Amtor and Pactor) overcome the fading problem on two way links but when there is more than one receiving station, as in a net or broadcast situation, a different approach is needed to get reliable copy. The MT63 mode overcomes selective fading by spreading the signal over a 1kHz bandwidth and using pretty powerful foward error correcting code (FEC). However, it does need a strong received signal level which makes it not really suitable for use on 3.5MHz during the day when path losses are likely to be high. It appears that there are two possible strategies for solving the problem which could be used either separately or in combination, viz:

1.  Spread the signal spectrally, either by generating a broad band of carriers (as in MT63 and DAB) or transmit several copies of the same signal in frequency multiplex using a standard modulation mode, as in our Combo test with three PSK31 signals. In the latter case, in addition to having simultaneous demodulation of the individual streams, it would be desirable to automate correlation of the received data streams and choose the best copy. This would be helped by some form of soft decision demodulation and it would probably make things easier if all character codes were of the same length, as in ASCII rather than Varicode. A problem with multi-carrier systems is the high peak to mean power ratio of the signal. To maintain linearity, on the rare occasions when all the individual sub-carriers add up in phase, they must not overdrive the transmitter or intermodulation distortion will occur. If the transmitter is set up so that the peak power when all sub-carriers add in phase is W watts, the peak power of each individual sub-carrier (assuming they are of equal amplitude) will be W * n-2 watts, where n is the number of sub-carriers used. If all the sub-carriers are of constant amplitude, the average power of the total signal is W/n watts. So, for example, a signal with 10 sub-carriers transmitted with a peak power of 100 watts would have a power of 1 watt in each sub-carrier and an average total power of 10 watts – not a very effective use of the transmitter’s capability. In practice, with many sub-carriers, the occurrence of the absolute peak is so rare that the transmitter can be somewhat over-driven without any serious intermodulation problem

2. An alternative to spreading the signal in frequency is to spread it in time. To overcome comparatively long fades of, say, 30 seconds, the bits of individual characters in the message could be coded with a fairly powerful FEC and then interleaved over a period of a couple of minutes or so. The receiver would then de-interleave the message and use the FEC coding to correct bits lost during any fades. Soft decision decoding and accurate long term timing would be necessary for the scheme to be effective. A decoding delay of two minutes would be unacceptable for normal ham communications but for broadcast purposes would probably be tolerable. Several variations on this theme are possible, e.g:

a. All characters could be sent twice, once directly and then repeated with long interleaving and FEC. This would give instant copy when the signal was good, with parts missed during fades filled in later.

b.  Simply repeat characters two or more times at specified intervals (e.g. 30 seconds and 80 seconds) rather than use FEC.

In all cases, it would be essential to have stable bit timing, so that the bits contributing to a particular character could be correlated even when there are intervening periods of fading. For this reason, it would seem that such schemes would be best suited to fixed character lengths rather than Varicode. A time spreading scheme has the advantages over spectral spreading that it occupies minimum bandwidth and the full power rating of the transmitter can be utilised.

A feature which appears to be lacking in nearly all amateur digital mode programs is some form of indication when a received character is missed or is of doubtful quality. This may be difficult to do without soft decision demodulation but would particularly useful when handling third party traffic which cannot be “interpreted” before being passed to the recipient.

Text (c) Ray Scrivens & Bernard Spencer 26/04/04.

Comments are closed.