Narrow band morse code with coherent detection

By Chris Turner, G7TZO*

An analysis of the baud rate of a 12 wpm morse transmission reveals that the baseband signal could be transmitted using a channel bandwidth of just 5 Hz. Every halving of the bandwidth reduces interference to adjacent channels and gives a 3dB signal to noise improvement to the DX receiver. Such narrow band operation might give morse code a 6 to 10 dB DX advantage over other digital modes such as PACTOR II. So why donít we do it?.


One of the reasons why narrow band is not commonly used is that the final detection of the signal is still usually performed by a human ear at an intermediate audio frequency around 750 Hz; the received signal never returns to a copy of the digital keying waveform that originated the transmission. The human ear detector has been trained on clean edged dits and dahs and presumably would find a low pass filtered version more difficult to decode. This objection can be overcome by demodulating the signal to its original keying waveform, detecting dits and dahs and regenerating clean versions of the pulses to modulate an audio tone for the benefit of the ear of the human operator.


Fig. 1 shows a block diagram design for a narrow band morse code transmitter. It is infeasible to implement narrow band filters at radio frequencies but fortunately an equivalent is to low pass filter at baseband prior to modulation provided that the modulation and the RF amplification are linear. This is the same requirement as for SSB amplification. Normal AM precautions, regarding the avoidance of over-modulation of the carrier, apply. Note that it is not enough to set the modulation depth using a continuous keydown since a filtered keying signal may overshoot the steady state value (and must not be clipped). The diagram shows nothing novel and many designs for each of the building blocks are available. The 5 Hz low pass filter may be an active RC 4-8 pole filter or a switched capacitor filter. It is desirable that its amplitude and phase characteristics become standardised as this aids the design of a matching filter at the receiver. Of course the AM modulator will generate two sidebands and so our RF signal will occupy 10 Hz bandwidth. This circuit would be quite easy to prototype for those that have a switched capacitor filter lying around and the resulting signal could be detected by ear on a standard receiver. Could you retrain your ear to decode narrow band morse?.


The block diagram for the receiver is shown in Figure 2. The standard CW or SSB receiver can be used unmodified since all the coherent detection takes place using the audio output signal

Assume that the standard receiver has been tuned such that the audio morse tone output is at 750 Hz. We may wish to employ an audio band pass filter centred on 750 Hz in order to attenuate any image noise at 1500 Hz which might get mixed into our synchronous detector. The audio signal is synchronously mixed down to baseband using a local oscillator also at 750 Hz. The IF is now DC or 0 Hz and so the main selective IF filtering is simply a 4-8 pole low pass filter with a 5 Hz bandwidth. Since the IF is 0, both sidebands of the AF tone will be mixed to the same overlapping baseband spectrum. If the phase of our local oscillator is just right we will get a useful maximum baseband signal. If the phase is 90 degrees off we will get cancellation and no signal. If the phase is 180 degrees off we get an inverted polarity baseband signal. To guarantee a detected signal, we therefore use two identical mixers and filters using the same local oscillator frequency but with a phase offset of 90 degrees between the local oscillator inputs of the two mixers. This results in two outputs which I have named I and Q standing for In-phase and Quadrature respectively. Now, regardless of the relative phase between the incoming audio tone and our local oscillator we will always have a signal on I or Q or both and, viewing the baseband signal as a phasor R, we have analysed it as the vector sum of the I vector and Q vector (see Figure 3.)

If we wish to measure the amplitude of R we can see from figure 3 that it will be the square root of the sum of the squares of the amplitudes of I and Q.

a = Ö (I2+Q2)

If we wish to measure the phase angle of the incoming audio tone with respect to our local oscillator it will be

f = arctan(Q/I)

If we wish to measure the frequency difference between the incoming tone and our local oscillator it will be the rate of change of the phase angle difference; when the phase angle stays constant then the frequencies must be identical. This same detector can be used to detect PSK, ASK, or any combination and is commonly used for GMSK detection needed by the GSM mobile telephones.

The I and Q outputs of the low pass filters are sampled by the decision and control logic block. One objective of this block is to adjust the VFO such that the phase angle difference between the local oscillator and the mixer signal inputs remains constant. This effects frequency locking to the 750 Hz incoming audio signal. The second objective of the decision and control logic block is to compute the amplitude of the signal vector R and decide if the signal is a dit or a dah or a space. The vector will be large in magnitude for dits and dahs and small or non existent for a space. When the logic has decided on the received symbol it regenerates a clean edged and correctly timed pulse which can be used to modulate an audio oscillator for the benefit of the ear of the human operator.



The hardware and software for implementing the decision and control block should not be overly complex. A cheap (£15) single chip 16 bit microcontroller such as the Siemens SAB80C166 has 10 analogue inputs with 10 bit ADC resolution and 8 pulse width modulated outputs which can implement analogue outputs at 16 bit resolution or local oscillator outputs. It has a fast (0.5m S) multiply and divide instruction for performing correlation or other simple DSP tasks. It can bootstrap its program from the serial port and (if needed) interface to external battery backed 8 bit CMOS memory chips. Most microcontrollers which have an ADC and an internal timer are suitable for this project since DSP capability is not essential. The two mixers and the 90 degree phase shift divider are available integrated as a single chip for GSM applications (e.g. Siemens PMB2402)


Does it work? I donít know!. Iím a new class B licensee and I am still learning morse. I do know that some morse reading programs seem to be very stupid (at least with my keying!) so perhaps that needs to solved first. I am currently trying to build an SAB80C166 microcontroller development system specialised for amateur radio applications.