Lyle,
I don't believe you can get something for nothing. The sample-in/sample-out
may work at steady state, but if you use an FIR filter, it still takes a
number of samples and time to fill the filter pipeline, as well as empty it.
My DSP may be a bit rusty, but I believe that the edges of a CW signal (the
no-signal to signal transition and the signal to no-signal transition) still
takes time to propogate through the FIR filter, and appears as a signal
delayed in time at the output. What is it about the K3 architecture that
eliminates this latency? How many taps does the FIR filter have that is
used in the K3 for a 30 WPM QSK mode, and what is the sample rate of the A/D
in that mode?
73,
Rich W1EZ
----- Original Message -----
From: "Lyle Johnson" <[EMAIL PROTECTED]>
To: <[email protected]>
Sent: Wednesday, October 10, 2007 2:13 AM
Subject: Re: [Elecraft] Comparison: K3 and SDR 5000A
As I understand it, latency is no longer an issue. Latency issues with
the early versions of PowerSDR and the SDR-1000 have been resolved. That
being said, signal processing latency is a fact of life for all digital
radios. We are talking microseconds,
I beg to differ.
Latency (delay) in a DSP-based radio (or SDR, if you prefer) is caused by
several things.
The one we can't get rid of is the filter delay. By making "shorter"
filters (fewer taps), we can reduce the delay through the filter, but at
the expense of filter performance. We get wider skirts, less ultimate
rejection, more passband ripple, or some combination of these three
factors.
However, DSP can be applied in many ways. One sure way to increase
latency is by processing the incoming signal in blocks. This means you
collect a certain number of samples, then process them all at once while
collecting the next block, etc. PC implementations of DSP typically use
this method.
Another method is to process the signal after each sample. This
eliminates the block delay. This is how the K3 processes signals.
As an example, let's consider an SDR using 2048-sample blocks (common in
the SDR world) and 96 kHz sampling. It will take (2048/96,000 =) 22
milliseconds to acquire this block. This is 22 ms more delay, or latency,
than a K3 will have, assuming similar delays in each radio for the
filter(s).
For a real world example, I connected an SDR-14 receiver (with associated
dual core 3 GHz PC) to the IF output of my K3. I then tuned in an SSB
signal and listened to it through the K3's speaker as well as demodulating
it and listening through the PC's audio system.
The audio coming from the PC was very noticeably delayed versus the audio
from the K3.
This delay or latency may not be an issue in all cases. But in come
common operational scenarios, like QSK CW or SSB using "syllabic" VOX, it
can be critical. By paying close attention to such latency issues in the
architecture of the K3, we are able to provide QSK CW operation at speeds
well over 30 WPM without "reducing the taps" in the filters or otherwise
compromising the performance of the radio. The 22 ms additional delay
cited in the example above would kill QSK performance.
Latency is just one of many considerations that arise when comparing
radios, architectures, and one's own needs and preferences.
73,
Lyle KK7P
_______________________________________________
Elecraft mailing list
Post to: [email protected]
You must be a subscriber to post to the list.
Subscriber Info (Addr. Change, sub, unsub etc.):
http://mailman.qth.net/mailman/listinfo/elecraft
Help: http://mailman.qth.net/subscribers.htm
Elecraft web page: http://www.elecraft.com
_______________________________________________
Elecraft mailing list
Post to: [email protected]
You must be a subscriber to post to the list.
Subscriber Info (Addr. Change, sub, unsub etc.):
http://mailman.qth.net/mailman/listinfo/elecraft
Help: http://mailman.qth.net/subscribers.htm
Elecraft web page: http://www.elecraft.com
