[RTTY] Analog vs software modems

[Kok Chen]

The following are just my opinions.  Those with different experiences  
will have different opinions.  So read this with a grain of salt.   
YMMV, etc.

Analog modems and software modems both attempt to approximate the  
solution of demodulating RTTY.

It is a matter of how close each one gets to solving the ideal  
equations.

In the case of modems, you also have to do it in real time.

30 years ago, you can only do rudimentary digital signal processing in  
software because machines that were fast enough are also very  
expensive.  I was micro-programming AMD bit slices and using TRW  
multiplier-accumulators back then just to process signals sampled at  
10 Ms/s.  Audio processing is much simpler of course, but even  
something like an 8th order audio filter which can be built for a few  
dollars using op-amps would have cost hundreds to build in the digital  
domain.

By the 1990s, desktop computers were fast enough (especially when you  
add the arithmetic processors to them; remember the 8087?) to start  
competing with the best analog modems.  Back then, Brian K6STI had  
already posted comments to this same reflector that there are things  
(such as AGC that "looks forward" in time, or the "digital flywheel")  
that are simply impractical to build with analog hardware. If I  
remember, Brian had also mentioned the fact that you can do non-linear  
implementations that would be impractical with hardware.

Approaching 2010, we now can do even more math on desktop computers  
than in the 1990s.  We can get much closer to implement the  
demodulation equations that we want by using software than by using  
the analog approach.  Even newer devices sold as hardware modems today  
are really DSP based modems (but they are much more limited in what  
they can do compared to the flexibility that you get from your desktop  
computer).

Sound cards also became better and cheaper between 1990 and 2009.

One of the problems with some software/DSP designs is that instead of  
starting at the basic first principals of demodulation, they were  
emulating what analog modems were doing, so they never were better  
than the analog modems.  As a result, software modems that are  
crippled that way fared even worse than analog modems once you go  
through poorly adjusted A/D converters.

To an engineer, hardware and software are merely parts of the set of  
tools.  With today's technology, low speed modems designed with  
software tools are incredibly more malleable and flexible than the  
hardware approach, not to mention the cost and the development time.

Imagine trying to build an Olivia modem without software/firmware.

True analog modems (not the DSP modems in a box) had two things going  
for them -- resolution/dynamic range and latency.

However, most sound cards today have dynamic ranges that already  
exceed the dynamic range of the transceivers they are used with.   
Plus, if you are using narrow I.F. filters, the dynamic range of a  
sound card does not even come into play until the signal is way weaker  
than the AGC threshold of the receiver -- and all you need to do in  
that case is make sure that the noise floor of the sound card is below  
the noise floor of the receiver.

Latency is also less of a problem as computers become even faster, and  
do you really care if the character the other person types appears on  
your screen 0.1 of a second later than if it were coming from an  
analog modem?

Latency is often a problem when dealing with real-time tuning  
indicators such as the crossed ellipse.  However, by transferring  
small sound card buffers and using decent sound protocols (ASIO in  
Windows and Core Audio in Mac OS X), even crossed ellipse lag should  
no longer be a problem -- if you see a significant tuning lag when  
using a crossed banana, it is a problem with the particular software  
implementation.

With the speed of a Core Duo machine, you can emulated every detail of  
the HAL ST-8000 analog processing in software to achieve the same  
performance.  Note that the ST-8000, with its analog AGC, has a  
dynamic range of 75 dB to 85 dB, something that most 16 bit sound  
cards can achieve *without* any analog AGC.  Emulating the KAM's  
filters and slicer is child's play.

But why would you want to emulate an analog modem when you can do much  
better with good software algorithms?

73
Chen, W7AY