[Laser] Optimizing sound card modes for optical communication

[TWOSIG at aol.com]

I have tried to point out in the past that amateur radio modes  using 
personal computer sound cards are not optimized for  optical communication.  While I 
think that the modes themselves have  great potential, they are optimized for 
use on radio rather than  optical channels, were limiting interference is of 
primary concern.   That usually manifests as limiting bandwidth, which in turn 
depends  on linear amplification and modulation processes.

The  experimental optical systems that are generally discussed here do not 
have  the interference problems of HF and frequently use components that are  
non-linear in their performance.  Sound card mode software tries to  produce 
clean sine waves.  Modified laser pointers work better with  square waves.  Said 
another way, analog modes may require some  adaptation to work on discrete 
(two state) systems.

Consider the  three digital modes:  PSK31, JASON, and MT63.  Here are some  
thoughts on how they should be optimized for transmission on optical  channels 
rather than RF.  These suggestions presume that the light  emitting devices 
are to be driven with square waves and that additional  harmonic energy does not 
interfere with the reception of the signals.   The overall conversion of 
energy from the power source into the desired  signal is not considered here, but 
I generally assume that the linear  amplifiers needed to preserve limited 
bandwidth are less efficient than the  switching devices that can be used to drive 
square  waves.



PSK31:

This is a very  narrow bandwidth mode.  The phase changes that take place to 
mark the  transition from ones to zeros and back take place at a time when the 
signal  envelope is near zero.  This limits any noise that might result from 
a  phase change that might take place at peaks, or even intermediate  levels.  
The sound card output is continuous, or should at least be in  fine 
increments.

To optimize PSK31 for optical channels, the output should be a  50/50 duty 
cycle square wave.  While strictly speaking it is not  necessary that phase 
transitions be limited to pulse edges, I recommend it  for BPSK.  The consequence 
is that the tone frequency is limited to  an integer multiple of half of the 
baud rate.  ( If I did the math  right, integer multiple of the baud rate would 
force whole cycles, that is  a high and low period to fit into the baud time, 
which was not intended.  Integer multiple of half the baud rate doubles the 
number  of frequencies available. )  Output is one of two stated with no  valid 
intervening value.  The spectrum will contain significant  harmonic content, 
but it is unlikely to affect signal decoding.   Choice of transmission 
frequency is not an issue on the transmission side  so long as you are sufficiently 
above double the baud rate.  It may  be that square waves do not perform as 
well as linear sine waves at  lower frequencies.  Operating frequency will 
probably depend on  receive system issues that I am not addressing here.

An  additional comment about the technique of using the sound card output 
with  amplification and clipping to "square" the signal.  The linear nature  of 
the PSK31 signals suggests that there will still be intermediate value  output 
near the zero crossing phase changes.  I suggest that a better  method is to 
use a threshhold ( such as a comparitor set at least half the  peak value ).  
The result will be square wave bursts with gaps between  them.  It should cause 
no degradation of the signal decoding as long  as the tone bursts are longer 
than the gaps.  ---- Perhaps this is  worth investigating the tone / gap ratio 
in low signal  conditions.



JASON:

JASON is a  narrow band frequency modulated system.  Its output is one of 16 
tones  with the data encoded as the difference in frequency between successive 
 tones.  If the frequency increment overflows the available tones, the  
system "wraps around".  The analog implimentation uses continuous  phase changes to 
limit transmit bandwidth.  Only four bits of data can  be delivered by the 
increments to 16 tones, so two increments are used to  encode one character.  
There is a possible confusion whether the  received increment is the first or 
second "half" of the character.   The system uses only eight increments for the 
first, and the other eight  for the second, resulting in no ambiguity, but 
only three bits of data per  increment.  The result is six bits per character, 
which is sufficient  for the communication system.

Optimization for an optical channel  for this system probably only needs 
square wave output.  As before I  recommend tone frequency change at pulse edges 
which limits the selection  of frequencies.  This in turn may cause the actual 
tone steps to be  slightly in error compared to the ideal.  However, the 
system was  designed to work with RF systems without "rock solid" frequency  
stability.  Therefore the errors in frequency increment should not be  a problem.

Operating frequency selection should be a receiving  equipment rather than 
transmit equipment  issue.




MT63:

MT63 uses  64 simultaneous tones, each BPSK modulated with forward error 
correction  coded data distributed in frequency and time.  Versions space the 
tones  over 500, 1000, and 2000 Hz, with the baud rate doubled in sucessive  
versions.  The intent was to provide reliable communication in RF  paths that have 
both frequency and time interference.

It is  not obvious that the characteristics of MT63 are suited to optical 
channel  issues.  However, the modifications that I suggest to adopt it  showcase 
a different way of thinking.  I suggest that the system be  modified so that 
each tone has the characteristics that I describe for  BPSK31.  Taking the 
1000 Hz bandwidth version at 10 baud:  Think  of the baud time as a frame.  Each 
tone should have a pulse edge,  rising or falling depending on the data, 
aligned with the frame edge.   The discrete frequencies that fullfill this 
requirement suggest that  frequency spacing between tones cannot be aligned as defined 
by the analog  parent.  The base frequency ( 500 Hz ) should be shifted 
upward so  that its third harmonic, which will be present in the square wave, falls 
 above the frequency of the 64th tone.  These two modifications should  
result in a modified system that will be broader ( perhaps 1300 Hz  bandwidth ) and 
starts at a higher base ( perhaps 700 Hz ).  These  transmit equipment issues 
should dominate the system design rather than  receive equipment issues.

This system also must be adopted for the  desireable two state transmission.  
Each of the 64 tones in the system  can have two states.  Since they are 
running simultaneously their  combined signal strength can vary from zero (all 
tones at zero state) to 64  (all tones at one state).  The output can be scaled, 
but it will exist  in, and can exist in only, 65 distinct states.  Sampling 
theory states  that you must sample at least twice the highest frequency 
component.   ( At the risk of being corrected by those more learned that I, I shall  
ignore the harmonic content of the square waves for this issue since  the 
harmonics are fixed in frequency and relative amplitude to  the fundamental 
frequency that I am attempting to sample. )

I  suggest that the system use a sample period that can be aligned with the  
frame ( baud period ).  For example, if I choose four samples and 2000  Hz as 
the high frequency to make the math easy, that results in 8,000  samples per 
second.  Once the input data is converted to the  64 simultaneous bit streams, 
the system can find a total for each  sample, which will have one of 65 
discrete values.  If I model my  system like a pulse density modulation system, I 
can use 64 clock periods  within each sample to repesent the values.   That 
allows me to  calculate 512,000 clock periods per second.  ( Or I could send out 7 
 bits of pulse coded modulation. However that is not the design I wish  to 
discuss here. ) 

Each sample's "on" bits can be sent  in any desired order.  If all "on" bits 
are clustered together the  result is pulse width modulation.  It will have a 
calculable frequency  spectrum for each value.  Interspersing the ones and 
zeros in various ways  will result in different, but still calculable, frequency 
spectrum.  I  have no idea which is more desireable, but as far as the 
transmission of  the data, they should all work.

This may seem to describe a  complex system, but the sampling rate and design 
limitations should be  "simpler" than an equivalent system that takes the 
analog audio output of a  sound card, encodes it for digital transmission, and 
then reconstructs the  signal for decoding on the receive  side.





MT63 may be an  extreme example of adaptation of a radio mode, whose 
complexity may serve  no real purpose in light communication.  However, the ideas can 
be  scaled.  If only 16 tones are to be sent, there are only 17 discrete  
values that need to be encoded.  Similarly, the use of resources may  show that 
sending single tones at higher baud rates are superior to systems  that encode 
multiple tones using a two state system.  Discrete systems  are better suited 
to optical channels where they do not need bandwidth  limitations.  Optical 
channels can probably work equally well for  analog channels, but the emission 
devices that I have assumed here do not  perform as well in analog systems.

I hope this will help  understanding the conversion of RF ( analog ) modes 
into Optical ( discrete  )  systems.


James
N5GUI