A common mistake in years past was to try to put all of the selectivity into
a single super-deluxe crystal, mechanical or digital filter. These filters
quite often have "raspy" noise interference at the edges of the passband
(especially in CW mode) due to the enhancements of spectral noise peaks at
the very sharp band edges, caused by the conversion of the phase statistics
of noise to amplitude statistics (each edge of the filter acts like a phase
disciminator). This raspy noise interferes with weak signals.

The filter can be equalized for group delay, as mentioned, with improvement
in the problem. The band edges can be softened, with good results. A much
better way is to use two or more intermediate-performance filters in
cascade. This method softens the edges so that the effect is greatly
reduced. It also improves overall shape factor. A cascade in this manner
of identical bandpass or audio lowpass filters tends in the limit toward the
Bessel or even the Gaussian response. Digital filters can also use a method
called Transition Band Sampling (see Oppenheim and Schafer 1975 or Oppenheim
and Willsky 1983). All of these results are related in principle to the
Central Limit Theorem of statistics.

The cascaded filter approach is also very beneficial in other respects, in
particular the reduction of wideband noise in high-gain IF amplifiers. This
noise degrades AGC performance and adds audio frequency noise to the product
detector output.

Bill W0IYH


"Dale Parfitt" wrote in message
news:b%X3f.301$W32.225@trnddc06...

Just sweeping a filter designed for SSB would be fine I suppose. I
recently had an opportunity to hear the difference between a stock FT-1000
CW filter and one homebrewed with attention paid to group delay- the
difference was very clear to hear- in favor of the homebrew filter.
\
Dale

"William E. Sabin" wrote in message
news:6J84f.437250$x96.418250@attbi_s72...

A couple of improvements in the following paragraph:

The filter can be equalized for group delay, as mentioned, with
improvement in the problem. The band edges can be softened, with good
results. A much better way is to use two or more intermediate-performance
filters in cascade. This method softens the edges so that the effect is
greatly reduced. It also improves overall shape factor. A cascade in
this manner of identical bandpass or audio lowpass filters tends in the
limit toward the Bessel or even the Gaussian response. Digital filters
can also use a method called Transition Band Sampling (see Oppenheim and
Schafer 1975)

Delete the incorrect second Reference to Oppenheim and Willsky 1983.

All of these results are related in principle to the Central Limit
Theorem of statistics.

See http://mathworld.wolfram.com/CentralLimitTheorem.html for this
interesting topic. The general idea that the theorem alludes to in this
example is that as sharp-cornered filters are cascaded the passband response
becomes noticeably more rounded at the corners, similar to Bessel and
Gaussian filters.

Smoothing and Windowing methods can be used to reduce sharp corners in
discrete sequences such as digital filters (see Oppenheim and Schafer 1975
and many other sources).

Bill W0IYH