Notes on Filtering & DSP

Contributed by George T. Baker, W5YR

There is a widely-held misconception that "a filter is a filter" and "you can't have too many of them." Nothing could be further from the truth . . .

Every conventional crystal lattice or ladder filter in the signal path contributes attenuation, varying response within the passband, relatively poor adjacent-frequency rejection (shape factor), comparatively poor ultimate rejection, substantial phase distortion at and near the passband "edge" frequencies, and non-trivial frequency-dependent time delay. A properly designed and implemented DSP filter has none of these deficiencies.

Some of these shortcomings impair noise-blanker operation. Effective noise blanking depends upon sensing the fast rise-time of the noise pulse before it propagates into the main signal channel, and switching off that channel until the pulse has passed. Every filter in the path deteriorates the pulse shape and degrades the rise time in addition to adding delay. The result is a noise blanker that doesn't blank well, due to noise pulse distortion attributed to the narrow filter.

Since the advent of DSP both for noise reduction and brick-wall filter generation, there has been another ill-conceived notion that conventional filters can be as "good" as DSP filters, provided only that enough money is spent. Wrong again, of course. No crystal filter can be constructed from a finite number of crystals that will equal the performance of a properly designed and applied DSP filter. The single drawback to DSP filters today is the inevitable time delay (group delay) which they introduce due to processing time. As processors get faster and cheaper, and that last IF moves to a few hundred kHz, even that minor disadvantage will fade away.

The need for narrow roofing filters vanishes if the system designer can ensure that from the antenna terminals to the output of the roofing filter - even at the usual 15 kHz bandwidth - and to the input of the A/D converter the operation of all circuits is linear, and that gains are controlled and distributed such that no signal or linear combination of signals can achieve an amplitude sufficient to overload the A/D converter.1 This is a challenging task but neither impossible nor unfeasible. The Ten-Tec RX-340, and the Icom IC-756Pro/Pro II, are real-world examples of "almost" meeting these design requirements. (Rockwell Collins gets even closer.)

Having met those non-trivial conditions, a narrow roofing filter is just another source of signal distortion. But, if you don't succeed in meeting those conditions, then narrow roofing filters are the easy way out of a poor design. If you can't prevent signals from distorting and cross-modulating due to non-linearities, then just accept only that narrow slice of the spectrum that contains the signal you want and try to reject the rest. Of course, the damage is already done in that the desired signal is now accompanied by distortion products, etc., but only those which lie within the narrow filter passband have a chance to proceed further and cause further deterioration, which they will.

As Adam VA7OJ/AB4OJ has said so many times, "provided that you do not overdrive the A/D converter," and as I have said so many times, "provided that you operate the front end in as linear a fashion as possible," then there is nothing that a narrow roofing filter can do that subsequent DSP filtering cannot do much better at significantly less cost. Take that back - the roofing filter can distort the signal in several ways which the DSP filter can not.

Bottom line: nothing today - and this was true back in the 50's when I studied Network Analysis and Synthesis under Ernie Guillemin at MIT grad school - in the conventional crystal filter universe can equal the capabilities of DSP filters. To use narrow crystal filters at all is a direct admission of an inferior front end design. Conversely, as the RX-340 or Pro/Pro II demonstrates every time you turn it on, narrow roofing filters are unneeded if the front end has been done "right."

We hear that Icom is taking the basic IC-781 design, with its quality/performance level still aimed primarily at the large and moneyed commercial and government/military markets, and updating it with what they have developed in the PRO series and taking advantage of the Rohde & Schwarz technology that they have licensed. With that combination they will set a new benchmark in the amateur market that will be hard to equal much less exceed. The radio is expected to be expensive - probably in excess of $5K or even $6K - but should be pack-leader by a country mile in the amateur HF marketplace.

IC-7800 Update: As discussed in the Icom publication "IC-7800 Technical News", the IC-7800 utilizes switched 15 kHz and 6 kHz 1st-IF roofing filters. While a 15kHz, or more, roofing filter is used in almost all current amateur transceivers, it is not the “optimal design” for SSB, CW, or AM. The IC-7800 utilizes two 1st IF roofing filters, one for FM operation and the second with a 6kHz bandwidth for SSB, CW, AM, and the Data modes. The effect of the 6 kHz filter is to reduce the broadband noise input to the ADC by 4 dB, as compared to the 15 kHz filter. This will improve close-in dynamic range in these modes.

1 "A High-Performance Digital Transceiver Design, Part 1" by James Scarlett KD7O, QEX, July/August 2002

2 "IC-756Pro II Receiver IMD and DSP Filter Performance",  by J. Saito JA7SSB, CQ Ham Radio, January 2002. Summary by N. Oba JA7UDE (PDF)

Copyright © 2002, George T. Baker, W5YR
Page created by A. Farson. Last updated: 06/16/2018