A MATCHED FILTER FOR EME

by Chuck MacCluer, W8MQW

In EME we know in advance that the signal is a (say) 500 Hz tone buried in noise. We wish to detect its presence or nonpresence. Is there a Morse code dit or dah being sent at this moment?

The best filter for detecting the presense of a known signal in white noise is a `matched filter,' where the received noisy signal is played against an uncorrupted version of itself --- much as an ideal local carrier is inserted for detecting SSB. The incoming noisy signal is convolved with say m=8 periods of a local clean 500 Hz signal. The improvement in signal-to-noise is a function of the receiver's noise structure. More on this below.

IMPLEMENTATION

Thanks to Texas Instruments one can easily and cheaply implement a matched filter. TI sells a $99 complete DSP board, the TMS320C5X DSP Starter Kit, referred to as the "DSK." The board is powered by any 9 VAC wall transformer @250 ma, has a serial DB9 port and two RCA audio in/out plugs. The levels at the analog ports are such that the DSK can be plugged directly between receiver audio output and headphones.

The board comes with two manuals, one a short overview, the other a 3 pound reference. The software is supplied on 3.5" media and runs on any MSDOS machine. TI ships a simple assembler, a loader, a powerful debugger, and several instructive tested routines, like for instance, a spectrum analyzer. You may write assembly source routines using any ASCII editor and file as file.ASM. Then invoke the assembler to produce strings of hex which the assembler stores as file.DSK. Once the DSK board is powered up with 9 VAC and once connected to the serial port, the loader will load and initiate execution of your routine. You may then unplug the serial cable. The DSK board will run standalone until depowered.

The DSK's onboard processor (TMS320C5X) is a 28.6 MIPS screamer with an 32 bit arithmetic logic unit, 32 bit accumulators, and a 16 bit address and data bus. The `Harvard' architecture has separate program and data memory. The assembly code uses 1--4 letter intuitive mnemonics, some of enormous power. The DSP chip was built to do convolutions with a vengeance. A simple matched filter requires only a tiny fraction of the capacity of the board.

The onboard analog interface circuit (AIC) does 14 bit A/D and D/A conversion at programmable rates from 7.2--19.2 kHz. It also provides anti-aliasing switched capacitor filters with programmable cutoff frequencies.

HARDWARE RECOMMENDATIONS.

Upon receiving your DSK board, immediately mount the board with conductive spacers to a metallic base, say plate aluminum or G10 PC board. This will ward off static and mechanical damage. If speaker level volume is needed, use an outboard amplifier like the $ 8 TenTek kit 1550.

Your nearest supplier of the DSK board can be obtained by calling TI at 800-477-8924. TI maintains a DSP BBS at 713-274-2323.

SOFTWARE

I have written a simple matched filter routine called MATCH.ASM. Once assembled and loaded it initializes the DSK to sample at 8 kS/s (8 kHz), using m = 8 periods of cosine for an improvement in S/N of 14 dB. The values of cosine are stored in tabular form in a separate file called COSINE.ASM. The assembler will include this table during assembly. I have posted MATCH.ASM and COSINE.ASM on the ARRL BBS and will ship via Email upon request.

For your requests: Chuck, W8MQV


The above are highlights of an article of the same name that appeared in the 1995 Proceedings of the Central States VHF Society. W8MQW


UNRESOLVED QUESTIONS

This investigation of the matched filter for use in EME has revealed surprising (to me) information about the noise structure of amateur receivers. I recorded one second of SSB noise from the speaker jack of a typical VHF receiver (IC271A) connected to a so-so preamp (ARR P144VD), sampling at 20 and 40 kHz, with and without AGC. I found the simulated improvement in S/N due to a matched filter (of a given number of periods) was independent of the center frequency of the filter. This flies in the face of theory --- in the presence of white noise, a matched filter centered at 500 Hz has 3 dB less noise reduction that one centered at 250 Hz.

After finally deciding that this conflict is not a mathematical artifact of causal sampling, I found that the "thermal" noise from the preamp+receiver is FLICKER not white noise; the data shows clear 1/f dependence on frequency, and this is NOT from the sampling process.

QUESTION: Is the noise from all amateur receivers flicker noise?

References claim modern devices have "noise corners," (where flicker noise changes to white), that are far below the audio frequencies where we listen to CW. But are the many semiconductors, multiple conversions and preamps of our systems combining to produce flicker noise well past 2.1 kHz?

SEND DATA

Can you digitally record a second's worth of galactic noise from your system? If so please ship me the data. I'll check its 1/f dependence.

If we are dealing with predominately flicker noise at audio there may be a more optimal noise reduction strategy.


For e-mail: Chuck MacCluer, W8MQV

maccluer@math.msu.edu
East Lansing, MI
8/15/95


For comments etc: W6/PA0ZN

For Top of Pages.