[Elecraft] DSP question

[Lyle Johnson]

Hello Vic!

> In Lyle's article on the design of the Elecraft DSP, he says "...with
> high speed CW the shorter delay may be beneficial."
>
> Lyle, could you expand on this?  What do you hear different as a result
> of more or less delay in the filter?

There are two aspects of the delay that are of interest: the actual delay,
and the time the signal "lives" in the filter.

Actual Delay

Here, we are interested in how long it takes before you hear the signal.  If
there is a delay of 50 mSec, for example, then it would take an additional
50 mSec after you have received the dit or dah before you begin to hear it.
This is the same as being another 9,300 miles away from the sending station
if you think of it in terms of propagation delay.

This may affect things if you are doing fast break-in.  Or it may not :-)

The other issue with delay is hearing your own sidetone.  In the case of an
external DSP unit, you have to run a keying line to bypass the DSP so your
sidetone isn't delayed.  Otherwise, your sync between key/paddle and ears
gets messed up really fast!

The KDSP2 doesn't have this problem due to its method of integration in the
radio audio path -- the sidetone bypasses the DSP completely and is injected
directly into the LM380 audio amplifier on the K2 control board.

Time in Filter

The type of filter used in most DSPs, including the KDSP2, is a so-called
FIR filter.  This type of filter can't oscillate,  and has many advantages
over feedback-type filters, such as common op-amp based filters.  The length
of the filter, or number of taps and the sampling rate all play a part in
determining how long a signal is in the filter.  This gives rise to effects
like ringing.

In designing the KDSP2, I set the filter lengths to very long and determined
how good the filter would be.  I then took into account real-world limits,
like "coefficient quantization" and looked at how short I could make the
filter and achieve the same practical result.

The net result is that the filters in the KDSP2 are shorter than found in
many external boxes, yet achieve the same real-world results.

In the KDSP2, the filters are 128 taps and sampled at a rate of 8 kHz, so
the "time to live" is 16 mSec.  The peak effect of the signal in the filter
occurs at 8 mSec, and would be normally specified as an 8 mSec delay.

73,

Lyle KK7P