Murray Greenman, ZL1BPU/ZL1EE, has an excellent discussion of this topic on the Fldigi site.
This is not intended to be a polemical discussion. There are lots of hams who are smarter than I am and far more knowledgeable than I am hold different views. That is probably because we value different facets of this complex hobby. But it may be because I am in error on some point.
This is also not intended to be a technical discussion of the communications theory but rather a discussion aimed at hams who are faced with making a decision about what mode they are going to use tonight to call CQ.
I enjoy conversational digital QSOs
In my years as a ham, I have found digital operators to be among the most sophisticated hams on the air. Many DX digital hams write good English. My favorite activity in this hobby is to have an extended digital QSO with a ham in a country with a different culture form my own where I have the opportunity to really get acquainted with the individual and learn to understand something about his values and personality. Much of my activities over the past twenty years have had as their goal to build a station that can hold a stable "circuit" for an hour with a marginal DX station during varying propagation.
Many modes are slow or else susceptible to ionospheric disturbances or QRM
Some digital modes excel in other types of QSOs. The quick "turn around" and tuning tolerance of 45 baud RTTY make it ideal for contesting. Dave, AA6YQ, has pointed out that the narrow bandwidth of PSK31 and the availability of panoramic reception software make it ideal for DX pileups where the DX station can use such software to pick out calls. PSK31 is also quite sensitive. Both of these modes have found widespread use with stations working toward DX awards.
But both of these modes are quite susceptible to ionospheric disturbances.. In response to this need, gifted engineers have developed a plethora of much more robust modes, some of which are exquisitely sensitive. In my opinion, some of these that are widely used for "ragchew" QSO's are too slow. Now many, if not most, young adult hams started typing in grade school and spending their valuable recreation hours limited to less than 30 words per minute is frustrating. During a recent QSO I turned up the volume on the radio, walked to the kitchen, made a cup of instant coffee in the microwave oven, returned to the radio desk, and read the other fellow's text before it was my turn to transmit. After one such QSO that lasted an hour I realized that we had exchanged so little information that I really did not know much about the DX ham. But fortunately, we have many modes to choose from.
I have graphed the bandwidth versus the speed of several digital modes above. I have chosen only modes with bandwidths of 500 hz or less only because it seems not to be good stewardship of the bands to use wider modes for ragchew QSOs. They may well be appropriate for file transfer, etc. Despite the intuitive bias that wider modes should have greater throughput, there is little correlation in the above graph. Why? The lack of correlation can be explained by the increased robustness of certain of the slower or wider modes.
I have indicated those modes that are slower than 30 words per minute, an arbitrary lower limit for keyboard QSOs that seems reasonable to me for most experienced typists. It is also useful to frame this broader discussion since there are modes both slower and faster with marked advantages for specific applications.
The data in the graph above is from Tony Bombardiere, K2MO, who used PathSim software from Moe Wheatley, AE4JY, to determine the weakest signal that would give good copy with each mode. Probably there would be little debate about this data. Other investigators might obtain values that differ by a db or so. Suffice it to say that there are several quite sensitive modes that are faster than 30 wpm and with bandwidths that are appropriate for the HF bands.
This data is also from K2MO. Here he used PathSim to introduce simulated ionospheric disturbances and measured the percentage of correct copy that resulted from passing text through each mode at various signal levels. I have grouped the data into three classifications used by Murray Greenman, ZL1BPU: Good Copy 90-100%, Difficult Copy 70-89%, and Impossible Copy below 70%.
I gathered some additional data
I wanted to confirm with my own hands data on those modes that would meet my requirements of being fast enough to keep up with typing while offering the possibility of being relatively robust with relatively narrow bandwidth.
- Recordings were produced with various digital software packages that testing had shown to be among the best for a given mode
- Audacity recorded the audio using Virtual Audio Cable running at 11.025 khz
- The text was a series of amateur calls. For those modes with larger character sets about two thirds of the text was lower case.
- Each data point represents at least seven minutes of text transmission.
- An attempt was made to avoid real time resampling of audio files. The 11 khz recordings were resampled in Audacity to 8 khz, 16 bit files using high quality sinc interpolation
- PathSim used the 8 hz files as input. The additive white Gaussian noise model was used to test sensitivity with the signal to noise ratio reported over a 3 khz bandwidth. To simulate poor propagation conditions PathSim used the CCIR 520-2 "Poor" parameters with two paths, 2 msec delay and 0.5 hz frequency spread in each channel. Noise was then added to these summed channels.
- The 8 Khz wave files produced by PathSim were resampled in Audacity to produce 11 khz files
- These 11 khz files were played by Audacity through Virtual Audio Cable to the decoding software
- RTTY was recorded with MMTTY 1.66G and decoded with TrueTTY 2.76.2 (UA9OV)
- MFSK was recorded with Stream and decoded with TrueTTY
- Contestia 500/16 was recorded and decoded in MixW2 2.16
- DominoEX8 was recorded in Multipsk 4.16 and decoded in Digital Master 780 5.0 Build 2434
- PSK31 was recorded and decoded with WinWarbler 6.6.2
- The received text was pasted into the editor of MorseCat (DK5CI) to remove non Baudot characters and then the text was pasted into RTTYCompare (VE3NEA) to compare the received text with the expected text.
Interpolating, I arrived at the following approximate signal to noise ratios for an error rate of 5%:
|Approximate SNR (db) for 5% Error|
The results did not confirm all of my biases. Although PSK31 is not as sensitive as some of the other modes, it performed better in this simulation that it it seems to perform on the air for me. Perhaps part of the reason I find PSK31 operation frustrating is because of its susceptibility to QRM.
These tests did confirm my experience with the excellent performance of MFSK16, my favorite mode. I find it interesting that under CCIR "Poor" conditions it performs almost as well as RTTY does with no ionospheric aberrations while consuming only modestly more bandwidth.
DominoEX was introduced to be simpler to tune and to not require transceivers of high stability. It uses incremental frequency keying where the data is represented by the frequency difference between one tone and the next. It occupies about the same bandwidth than MFSK16. It did not meet my expectations under the "Poor" conditions. I found it difficult to get some software implementations of DominoEX8 to perform well, perhaps because of my inexperience. One could, of course, switch to DominoEX8 FEC but that would not meet my requirements for approximately 40 wpm. Contestia 500/16 performs better than DominoEX8 under poor conditions but consumes more bandwidth.
I was surprised by how much the implementation of some of these modes differed in different software. A given software program would have an excellent implementation of one mode and then have other modes which were not competitive in my hands. If you don't have luck using a given mode, don't give up on it before trying it with another software package.
Path simulation software does not model all the insults and vagaries of HF propagation that our signals meet on their way around the globe and they certainly don't indicate how modes deal with QRM. But the data is useful in making suggestions about how robust these various modes are.
So which digital mode should I use tonight for a ragchew QSO?
I am quite agnostic on this question. I am not certain which is the best mode, and that may change as new modes or new software for existing modes are developed. In designing any mode an engineer makes a decision about how to balance throughput and bandwidth to arrive at a desired level of robustness. But from this data it appears that MFSK16 and Contestia 500/16 offer an appropriate balance of these constraints for extended conversational QSOs with changing propagation conditions.
But I can't even recognize those modes on the air!
As of November 2009 Patrick Lindecker, F6CTE has developed an answer to this issue. Several digital software packages include his Reed Solomon ID protocol that sends a two second code before each text transmission informing the software at the other end what mode is being used to send the message and the center frequency for the transmission. At this time I am aware that Fldigi, Digital Master 780 v5, and MultiPSK all implement this protocol in their latest versions. Recent versions of all three of these packages also include the two suggested modes. For MixW users to obtain Olivia, Contestia, and RTTYM it is necessary to download the dll files to their MixW subdirectory.
Of these three, MFSK16 uses less bandwidth but hams often complain that it is difficult to tune. In response to this impediment, recent versions of some of the software packages lengthen the lowest tone at the beginning and the end of the transmission as a tuning aid. It is interesting that MFSK16 and Olivia are multiple frequency shift keying modes rather than phase shift keying modes even though it is faster to detect the phase of a signal than the frequency of a signal. One advantage of frequency shift is that if only one tone is transmitted at a time the transmitter is not required to be linear. Also with a single tone the transmitted power is not spread over several frequencies.
So I don't care what mode you use tonight. Just get on the air and if one doesn't work, we can change.