OpenBCM V1.13 (Linux)

Packet Radio Mailbox

DB0FHN

[JN59NK Nuernberg]

 Login: GUEST





  
G8MNY  > TECH     14.06.07 07:40l 150 Lines 7144 Bytes #999 (0) @ WW
BID : 18905_GB7CIP
Read: GUEST OE7FMI
Subj: Modem Bauds & Bits
Path: DB0FHN<DB0MRW<OK0PKL<DB0GRL<SR6BBA<SR7DWI<SR1BSZ<IW2OAZ<IK2XDE<DB0RES<
      ON0AR<GB7CIP
Sent: 070613/2305Z @:GB7CIP.#32.GBR.EU #:18905 [Caterham] $:18905_GB7CIP
From: G8MNY@GB7CIP.#32.GBR.EU
To  : TECH@WW

By G8MNY                               (New June 2007)
(8 Bit Graphics use Code page 437 or 850)
Do you know the difference between baud & data rate. BAUD is the changes per
second, & DATA RATE is the number of bits per second, say leaving the computer.
In simple AFSK or FSK there is a 1:1 relationship. eg. 1 bit = 1 baud.

FSK 1200B/S
                                  This is fairly typical of what happens as
             0         1          the FM idle carrier at 1800Hz causes 2 AM CW 
 ³    600Hz /³\       /³\ 600Hz   signals to appear at the deviation set points
 ³     of  / ³ \     / ³ \  of    representing the data 0 & 1.
 ³    LSB /  ³  \   /  ³  \ USB   Each of these carries full double AM side-
 ³       /   ³   \ /   ³   \      bands up to the baud rate/2 in Hz, as it
 ÅÄÄÄÄÄÄÁÄÄÄÄÁÄÄÄÄÄÄÄÄÄÁÄÄÄÄÁÄ    needs 2 bits to form a cycle.
 0 300 600  1k2  1k8  2k4   3K    (1800Hz carrier is used as it is the fastest)
                                  (frequency on most telephone lines.)

For this FSK to work, a bandwidth of 600Hz - 3kHz is needed, it does not have
to be too level flat, as FM can be hard limited & remove most noise etc.

So a bad line/comms link with more than 15dB frequency spread & worse than 15dB
S/N is still error free despite being very poor for phone use.

However differing delay times in the 2 tones is critical, & a Group Delay
difference of less that 1mS between 1200Hz & 2400Hz will garble the data, as
the modem will Rx no tone, then 2 tones etc. This has no affect on comms audio!

Typical long loaded line, or via a comms Rx IF filter!
   Delay
 1200uS´ \ Transformers              / Channel
  900uS´  \  & caps                 /  Low pass
  600uS´    \                      /  Filtering
  300uS´      ~~ÄÄ..__         _.Ä~
    0uS´              ~~ÄÄ---Ä~
       ÅÄÂÄÄÂÄÄÄÂÄÄÄÄÂÄÄÄÄÂÄÄÄÄÂÄÄÄÄÂÄÄÄÂÄÄÄÂÄÄÂÄÄÂÄÄ>Frequency
       300 500 800  1k2  1k6  2k   2k4 2k7 3k 3k2 3k4  Hz

Sent data                                Rx Data with 833uS of group delay!

______~~~~~~____________~~~~~~~~~~~~    ______????????????______??????~~~~~~

  0     1     0     0     1     1         0     no   0+1     1    no     1
1200  2400  1200  1200  2400  2400      1200   Sig 1k2+2k4       Sig   2400

This still is one of the simplest & robust standards around (this why it is
still used), but even so it is important to have all the sidebands that make up
the signal turning up all at the same time & at the same relative level & also
not drowned out with noise or distortions.

GOING FASTER
If we look at what a telecomms line can pass eg. 300Hz - 3.3KHz with some
frequency attenuation & group delay. Then that would support a max baud rate
of 3000 baud, eg. an 1800Hz carrier with double sidebands up to ñ1500Hz.

Alternating baud signals of 010101 @ 3000 baud form the highest side bands of
ñ1500Hz.

 ³ 1500Hz   /³\
 ³  of    /  ³  \  1500Hz
 ³  LSB /    ³    \  of
 ³    /      ³      \ USB
 ÅÄÄÄÁÄÄÄÄÄÄÄÁÄÄÄÄÄÄÄÁÄÄÄ> Frequency
 0  300     1k8      3k3   Hz

However this is not the data rate, but the max baud rate. So by using AM with
multiple levels & phase modulation with multiple angles of carrier together, it
is possible to make a single baud of carrier represent 1 of many states, eg. 16
AM levels & 16 phase angles, gives 256 states or symbols, which is equal to 8
bits of data per baud! That would make a data rate of 8x 3k = 24kB/S.

Phase & level diagram (dartboard vector) for just 4x 4 (16 symbols)

  +4´ [0]                     [1]
  +3´    [4]               [5]     This give 16 pigeon holes (0-F) for each
  +2´       [8]         [9]        burst of carrier signal (baud) to drop into.
  +1´          [C]   [D]
   0´              0               In practice the 4 phase are rotated per
  -1´          [F]   [E]           level shell to give the biggest pigeon
  -2´       [B]         [A]        holes.
  -3´    [7]               [6]
  -4´ [3]                     [2]  Note that the carrier level of 0 can't be
    ÀÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄ  uses as it would not give any phase info.
LEVELS-4 -3 -2 -1  0 +1 +2 +3 +4


  +4´ [ ][ ][ ][ ]   [ ][ ][ ][ ]
  +3´ [ ][ ][ ][ ]   [ ][ ][ ][ ]   By using variable number of phases per
  +2´ [ ][ ][ ][ ]   [ ][ ][ ][ ]   carrier level, you can fill all the
  +1´ [ ][ ][ ][ ]   [ ][ ][ ][ ]   pigeon holes as close as possible, giving
   0´              0                the biggest number of symbols, here 64,
  -1´ [ ][ ][ ][ ]   [ ][ ][ ][ ]   or 6 bits per Baud.
  -2´ [ ][ ][ ][ ]   [ ][ ][ ][ ]   
  -3´ [ ][ ][ ][ ]   [ ][ ][ ][ ]   Noise, distortion, poor frequency/phase
  -4´ [ ][ ][ ][ ]   [ ][ ][ ][ ]   response will move the carrier into the
    ÀÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄÄÂÄ   wrong pigeon hole, & produce 6 errors
LEVELS-4 -3 -2 -1  0 +1 +2 +3 +4    at once!

However you don't get something for nothing, using 16 levels means the
distortion & all noise must put the level into the next 6.25% level window, or
3.1% (-31dB peak). And the same goes for multiple phase mod, due to jitter &
multipath etc. that must not to exceed 1/32 of a cycle peak @ 1800Hz! A very
tight specification for radio path!

EQUALISING (training)
As the line will not be flat or have no group delay it is important that the
modem has an equaliser that has the opposite characteristics to the line. This
used to be an analogue nightmare, but with digital A-D convention of the line
audio it is relatively simple to flatten in software!

eg.. a Line with bump at 1K
Level                          Add in % of signal data from 500mS ago & the
  ³            /~\             bump will go. But then you get a bump @ 2kHz,
  ³     .-----~   ~----.       so add in some signal data from 250mS ago, keep
  ³    /                \      doing this until all bumps are above 3.4kHz.
  ÅÄÄÄÂÄÄÄÄÂÄÄÄÄÂÄÄÄÄÂÄÄÄÄÂÄÄ
  0  250  500  1k   2k    4k

Training can also be done on live data, by making small adjustments & seeing if
the data quality (how near Rx symbols are to their pigeon hole ideal centres)

  Ideal       Amplitude     Phase       Noise &
Pigeon hole    Trouble     Trouble    Distortion
ÚÄÄÄÄÄÄÄ¿
³       ³         ±                      ±±±±
³   o   ³         ±         ±±±±±       ±±±±±±
³       ³         ±                      ±±±±
ÀÄÄÄÄÄÄÄÙ       zero         zero

So you see that although modern modems are super at equalising & giving 56kB/S
max on a good line short, they can only do that over clean copper circuits, &
high quality PCM systems that give <1% THD >50dB S\N & no phase jitter.
In practice most lines can't do this the modem is designed to drop back the
baud, & or levels/phase angles (called symbols rate). Also the modem has error
detection/correction so missed errors passed through are low. The use of an
encryption algorithm in some modems also minimises continuous retries of data
patterns that have high error rates.

BROADBAND
These systems are different & use much wider frequency range & multiple
carriers. But there are similarities & nasty line problems will greatly reduce
the usable data rate.


Why Don't U send an interesting bul?

73 de John G8MNY @ GB7CIP


Read previous mail | Read next mail


 29.01.2026 21:04:09lGo back Go up