Duplex Radio-Telephone (Revisited)

Main What is super-regeneration? W. B. Lewis Back to projects

 1936  2008


In September 1936, The Wireless Engineer published an article by Lewis and Milner describing full-duplex radio-telephony apparatus using only two valves! A short summary appears in "Super-Regenerative Receivers" by J. R. Whitehead (Cambridge University Press, 1950) where I first encountered it. I am indebted to W. E. Legg, the curator of HMS Collingwood Royal Naval Museum of Radar and Communications, who supplied a photocopy of the original article from which I created this PDF:

  A Portable Duplex Radio-Telephone (450 KB)

Below are some observations on Lewis and Milner's work and an account of my efforts to replicate it. I tried replacing their two valves with a pair of transistors; but my first prototype lacked stability; and I opted instead for a digital PLL-based approach which seemed a safer bet. Radiating a peak power of only 5 dBm using horizontally polarised folded dipoles 1.8 metres above ground, high quality audio with very little background hiss was exchanged over a distance of 100 metres. This is not the maximum range, but the length of the OFCOM-approved test site. A Test and Development Licence was obtained from OFCOM, the UK licensing authority, before conducting a field trial.

Range and locking

L&M communicated over ranges from below 200 yards to tens of miles using quench frequencies between 50 and 100 KHz. Under the heading "Mode of Operation of the Modulated Quench Duplex System" on page 478, ranges and suitable quench rates are listed as follows:

"up to 200 yards""any frequency might be used"
"200 yards to ¼ mile""only lower frequencies could be used"
"¼ to ½ mile""a high frequency is most suitable"
"still greater distances""correct frequency found by trial"

The last paragraph on page 478 begins "When the theory of operation is considered more exactly, it will be realised that it is necessary for a modulated part of the pulse from each set to reach the other in its sensitive phase." That locking may "occur either at the beginning or end of the pulses received" is only mentioned once, in passing, on page 479. L&M only partially explain the locking mechanism. An example is given where both sets lock on the beginning of the pulses; but this only works over ranges from 1.5 to 3 kilometres. When working longer distances, more than one pulse is in flight between the sets at a time. The mechanism for working shorter ranges is less obvious. Three locking schemes are illustrated below:

Leading-edge

Asymmetric (both edges)

Trailing-edge

For a given quench frequency, asymmetric and trailing-edge locking work over shorter ranges than leading-edge; however, trailing requires adjustable pulse width, over which L&M had no direct control; it also seems impossible with a simple two-valve circuit for another reason: if a pulse from set A is sampled too early by set B, it is sampled at a point of greater amplitude, local build-up at B is accelerated and the next pulse is even earlier. B pulses arrive earlier at set A, which samples at a point of lower amplitude, causing A to slow down!

Coupling between antenna and oscillator affects pulse width. Heavier loading stops oscillation over a greater portion of the quench cycle. Quench amplitude is another variable. The active device creates negative resistance which counteracts losses and loading on the circuit. It's a delicate balance. L&M may even have controlled the relative phases of their quench oscillators by slightly offsetting the frequencies. Coupled oscillators lock with a phase-offset if their natural frequencies are not identical. I suspect much skill was required to operate the sets!

L&M probably used asymmetric locking for ranges up ½ mile and leading-edge beyond; but it's unclear why "only lower frequencies could be used" between "200 yards to ¼ mile" whilst "a high frequency is most suitable" from "¼ to ½ mile." It's a pity they didn't say more about locking mechanisms and operating procedures. Filling eight pages without these details, perhaps the article was constrained by editorial space? I would love to hear from anyone who has more information on this or other early duplex systems.

First attempt

Despite uncertainty about how it would work - an unsound basis on which to proceed - I tried to replicate L&M's results using a simple two-transistor circuit based on their valve original. Results were disappointing. I achieved quench-lock; but audio was either one-way only (half-duplex) or badly distorted. I'm sure it could be made to work, with enough patience; but I decided to abandon it and try a more sophisticated circuit using digital envelope detection and phase-lock techniques. This was my first attempt for the record:

VHF oscillator Q1 and quench oscillator Q2 were coupled through C3 and L6. I hoped dual-gate MOSFETs would exhibit valve-like properties! Audio was extracted at the junction of R9, L1 and C11. Modulation was applied to the gates of Q2 through a 1M resistor. Transformer L4/L5 was a Toko 7mm 5306. L2/L3 were 1.5 turns each of 20 SWG enameled copper wire, closely coupled. L1 and L6 were 6 turns of the same, mounted mutually perpendicular. V7, C1, C2 and C5 were variable.

Super-regenerative oscillator (SRO)

This super-regenerative oscillator (SRO) has a digital interface:

Q1 is a common-gate Colpitts oscillator. Coaxial resonator L2 gives better stability and Q than wound inductors at VHF. Trimmer VC1 primarily adjusts antenna loading whilst VC2 adjusts frequency with some interaction.

With the input capacitance of CD4007 U1A thru D at pins 3 and 6, schottky diodes D1 and D2 form a peak detector. Two diodes seemed optimal in Spice. The BAT81 has a diode capacitance of only 1.6 pF, which minimises loading. Q1 drain is at +9V when oscillation is quenched. PMOS transistors U1A and C conduct when the oscillatory voltage exceeds a threshold. U1E and F speed-up transitions, buffer and level-shift the DETECT output to 5V logic.

Driven directly from 5V logic, R3, R4 and C7 control quench shape and amplitude. With the values shown, super-regeneration is step rather than slope-controlled. The R/C ratio could probably be improved. R3+R4 is quite a load for 4000-series CMOS.

Delay between the rising edge of QUENCH and the falling edge of DETECT depends on received signal amplitude, noise and the setting of VC1. Triggering the 'scope off QUENCH, in the presence of a strong signal, the DETECT edge is sharp and early. With noise alone, it's late and jittery. Edge position exhibits a logarithmic response to signal strength.

C2 and C6 are quite large to mask variations in Q1 capacitance over the quench cycle which make the frequency of maximum emissions slightly higher than that to which the circuit is most sensitive as a receiver. L&M reported a similar problem with valves!

The metal-canned 40673 or 3N211 work well at Q1. Both can easily handle the power dissipation which is a problem with this circuit. Peak output is around 5 - 10 dBm, depending on loading; however, more power is wasted through dissipation in Q1. In fact, like a valve circuit, it takes a few seconds to warm-up before the operating point stabilises!

Quench PLL

Here is a block diagram of one transceiver:

The communicating sets are master and slave. The master has a fixed quench frequency to which the slave locks. Trailing-edge locking is used. The slave keeps its sensitive instant in the middle of the trailing-edge at the end of received pulses. The lock monostable sets exactly where on the trailing-edge the slave tracks. It incorrectly locks on leading edges if the PD inputs are reversed.

Transmitted pulses are slightly width-modulated. PLL loop bandwidth is narrow (about 300 Hz) to prevent it tracking and thus removing audio modulation. Narrow bandwidth also aids noise rejection. Received audio is extracted at the PD output. LPF1 bandwidth is 5 KHz. Very little local sidetone is heard and loopback testing is possible, since the PD only responds to rising edges which do not carry the outbound modulation.

The width monostables are adjusted for clear received audio at the master, placing the modulated edge at its sensitive instant. The master's lock monostable sets the mean level of its PD output. This has no effect on AC amplitude of recovered audio; however, there is a puzzling "dead-band" where the audio suffers crossover distortion unless the PD output is slightly offset away from zero phase error. Despite tracking with zero phase error, the slave's audio is not so distorted.

This 'scope screenshot shows QUENCH and DETECT with the sets deliberately prevented from locking. The local and remote quench oscillators differ slightly in frequency and are sliding relative to one another in phase. Two distinct edges are evident on DETECT as the sensitive instant falls either on or between pulses. Note the greater jitter on the second of the two:


HORZ 1µs/div

Oscillations quench faster than it would appear - falling DETECT edges propagate slower than rising edges through the SRO CD4007.

If the PLL does not capture by itself, the VCO can be swept onto frequency by retarding and then advancing the LOCK edge. This works because the the average PD output depends on where the LOCK monostable timeout occurs between the two DETECT extremes.

Modem

Modem (modulator/demodulator) seems an appropriate name for this board:

The 4046 VCO is disabled and phase detector PC2 is the only part of U3 used. PC2 is a phase and frequency detector (PFD) but only detects phase here, since pulses arrive with the same frequency at pins 3 and 14 due to the sampling nature of the quenched receiver. Capture range is limited by the necessarily narrow loop bandwidth - a characteristic normally associated only with XOR gate PC1. Unlike PC1, however, PC2 locks with zero phase error.

LC VCO U2/D has a narrow tuning range and stable centre frequency. KVCO is approximately 50 Hz/V. C11 and C14 are high-quality silver-mica types and transformer L1 is a Toko 7mm 5306 coil with Q=55. R12 prevents U2/D output loading the tuned circuit, maximising the operating Q. I wonder if adequate stability could be obtained from the 4046 VCO?

U2/C is an active loop filter which holds U2 pin 5 at half-rail. Jumper JP1, which is fitted on the master only, clamps the output to half-rail. If both sets are configured as slaves, they lock to one another and then drift up or down together, until one VCO control voltage reaches GND or +5V.

Audio amplifier U4 is preceded by a two-pole filter which removes the 100 KHz quench frequency. Audio can be taken from U2 pin 4 with local sidetone; but is best taken without sidetone from U3 pin 13. U2/C keeps this node at half-rail when the phase detector output is tri-state.

U2/F is the microphone amplifier which applies a slight modulation to pulse width monostable U1/B. The other half of the CD4098, U1/A is the lock monostable. Both are triggered by the same VCO edge.

Spectra

Spectrally, super-regenerative oscillation resembles lopsided direct sequence spread spectrum. The oscillation depicted on the left is building-up from noise alone. That on the right is locked to another signal - hence the quieting - and energy is concentrated in tones spaced at the quench frequency.

Lewis and Milner used a wavelength of 3 metres (100 MHz) which today falls in the highly congested VHF FM broadcast band. Modern wideband and ultra-wide-band (UWB) modulation schemes are generally confined to microwave frequencies, although some work has been reported in the 30 - 50 MHz range. And papers have been published recently describing full-duplex UWB transceivers using super-regenerative oscillators!

Modulation depth

Modulation should be "very slight" according to Lewis and Milner. I used less than 5% in my experiments. Nevertheless, recovered audio fidelity is good and signal-to-noise ratio is high. How can this be?

OFCOM Test and Development Licence

I only needed to briefly radiate a few milliwatts close to ground level in a built-up (i.e. well-screened) area; but I feared OFCOM might object to such a wideband emission or demand test equipment calibration certificates! In the event, my application was approved without fuss.

The application form and guidance notes on its completion can be found in the non-operational licensing section of OFCOM's website. Non-operational means temporary. These licences are typically granted to academic researchers and companies developing commercial radio products; but anyone can apply. OFCOM deserve credit from the Plain English Campaign for making the form and associated documentation so straight-forward. Approval takes about 6 weeks and a one-month licence costs 50.

OFCOM require a class of emissions using the modern International Telecommunications Union code. I settled on M8EJT meaning:

M = Modulated in position/phase
8 = Two or more channels containing analogue information
E = Telephony (including sound broadcasting)
J = Sound of commercial quality
T = Time-division multiplex

I somewhat arbitrarily stated bandwidth as 2 MHz with a power spectral density of below -75 dBm/Hz at offsets over 1 MHz.