Re: [hylafax-users] ECM overhead (was: modem selection)

[Lee Howard <faxguy@xxxxxxxxxxxxxxxx>]

On 2004.03.15 09:37 Sebastian Flothow wrote:
> Am Montag, den 15. März 2004, um 04:04, schrieb Lee Howard:
> > ECM faxing will still generally take longer than non-ECM faxing.
>
> I recently read the manual of one of these all-in-one machines (fax, 
> scanner, printer, copier) made by HP, and it stated explicitly that 
> ECM faxing does not take any longer than non-ECM faxing.

This is not true except in some rare circumstances.  However, I suspect 
that they were either only considering the negotiated bitrate transfers 
(14400 bps), they consider the time difference between a 40-second fax 
and a 45-second fax to be negligible, or they're trying to factor in 
MMR compression savings somehow, or some other weird statistical 
analysis.

The truth is that the image data is transmitted/received with ECM in an 
entirely different method than without ECM.  So they don't compare 
easily enough to say categorically "they take the same amount of time" 
or "ECM is slower than non-ECM".  One must say "usually", "roughly", 
and "generally".  Instead of transmitting the entire page of image data 
at once and then asking if the quality was acceptable or not (as is 
without ECM), with ECM up to 64KB of image data is transmitted as a 
"block" with each block divided up into as many as 256 "frames" which 
each contain a 16-bit FCS/CRC sequence and other framing bits and then 
the receiver is asked if all frames in the block were received 
perfectly or not.

So the fundamental difference between ECM protocol and non-ECM protocol 
is that ECM expects *perfect* image quality while non-ECM only expects 
*acceptable* image quality.  So the major difference in time between 
them occurs when there is an expected small amount of image data 
distortion.  If one or two frames are bad then ECM will require a 
retransmission of those frames while they will be accepted as-is 
without ECM.

Even if we assume that there is no image data corruption occuring and 
that each page of image data is less than 64KB, the amount of data that 
must be transmitted is still slightly higher with ECM due to the 
synchronization sequence and additional framing bits.  This difference 
can generally be considered negligible, however.

The only time non-ECM will take longer than ECM is when there is enough 
image data corruption to trigger a rejection by the receiver while the 
comparable ECM session would get the entire image more quickly 
piecemeal.  This is a rare circumstance, but it does occur.  Usually a 
receiver or a transmitter will give up (disconnect) sooner on non-ECM 
retransmissions than they will with ECM.  So with ECM most equipment 
will tolerate a dozen retransmission attempts on a single block (per 
spec), but without ECM most equipment will only tolerate three 
retransmissions of a single page (per T.30 spec).

> Of course, the manual was pretty much dumbed down, and I don't really 
> believe this claim, but the obvious contradiction with what Lee (and 
> maybe others) said here sparked my interest.
>
> So, how much overhead does ECM actually cause?

Just for an example I've done two single-page fax scenarios on clean 
lines (no telco, no modulator, no wires, no noise, no retransmissions 
via software: t38modem-to-t38modem).  Times are calculated as the time 
on the receiver from ATA to ATH.  The image uses fine resolution, A4 
page sizing, is 2D-MR compressed, and is transmitted at 14400 bps 
(should be a typical and fair comparison).

First: faxing the default HylaFAX coversheet (one block)
       ECM : 26.33 sec.
   non-ECM : 25.81 sec.
difference : 2%

Second: faxing /usr/share/ghostscript/7.03/examples/escher.ps (multiple 
blocks)
       ECM : 190.92 sec.
   non-ECM : 167.59 sec.
difference : 12%

> I assume it depends strongly on line quality, as a poor line will 
> require more retransmits.

Yes, that's true.  It also depends on the image size.

> Will it cause overhead on a perfect line, where no retransmissions 
> are necessary?

Yes, but 2% - 12% may be negligible to you - it apparently is to HP.  
PLUS, you get the added benefit of "guaranteed" perfect image quality.  
So the cost of perfect image quality is that 2% - 12% time increase.

> And, how big are the chunks which get retransmitted?

In ECM the only chunks that are retransmitted are the frames (usually 
256 bytes each) that were corrupt.  In non-ECM the whole page must be 
retransmitted.  So you can see that ECM will always be slower until 
enough data is corrupt to trigger a non-ECM retransmission.  At that 
point ECM will could be (and likely will be) quicker.  Thus, if you 
have constantly noisy lines then ECM may actually save you time in 
general... but I think that will be infrequent.

But again, the comparison between ECM and non-ECM really isn't with 
time... it's with expected image quality.  So comparing them on a time 
scale isn't really fair unless you understand that principle.

Lee.

____________________ HylaFAX(tm) Users Mailing List _______________________
 To subscribe/unsubscribe, click http://lists.hylafax.org/cgi-bin/lsg2.cgi
On UNIX: mail -s unsubscribe hylafax-users-request@xxxxxxxxxxx < /dev/null
 *To learn about commercial HylaFAX(tm) support, mail sales@xxxxxxxxxxxx*