AD630 recovering morse signal 100dB below antenna noise

Thread Starter


Joined Mar 24, 2012
Since this doesn't seem to work as expected from my experiments, I'm throwing it open to discussion rather than keeping this Limbricks' Enigma idea under wraps as some sort of 'stealth' transmission mode.
All the following are with a 2v peak RF carrier with just a dummy antenna resistor, so technically not transmitting.

I chose to use a 7.023 MHz carrier and AM modulated with a 32.768 kHz sine wave from a watch crystal oscillator, gated at 5 Hz (20wpm).

I used another watch crystal oscillator as the reference.

Why 32 kHz? The experiments with watch crystals confirmed research on-line, these can be ultra stable. Several days with a only a few degrees of phase shift at identical temperatures between a pair of crystals, extremely low jitter. The Q of these tuning fork crystals is incredibly high (they will carry on ringing for several seconds after initiated by another crystal oscillating nearby). The exact frequency is irrelevant, the need is for a reference, not a time keeper.
If pre-determined drift on receive of the reference is intended until some part of the signal is picked up, a PLL can take over, like a key in a lock.
As far as extracting the AM from the carrier, I tried a Direct Conversion Receiver (my own ultra low noise homebrew). Instead of a CW tone at say 700 Hz, it is 32 kHz.
The AD630 outputs the 5 Hz when buried in antenna noise, but only to about the same threshold as the ear can just detect anyway(when receiver retuned to audible beat) . Not very stealthy then!
I tried a high Q bandpass before the AD630. It didn't help (obviously the lock-in amplifier is very high Q effectively, as it's working principle). What I did see on the scope of this filtered signal was other passing 32 kHz signals. Harmonics of SSB? CW?.
These are the only explanation for the random 'blips' which prevent higher gain of the low pass AD630 output.
Antenna noise is not 'white noise'. Which I have always suspected, looking at spectrum plots. Some significant part must be signals overlapping in a mess of 'straw'. Certainly, there is a dip in spectrum where SSB transmissions are filtered below 300 Hz.
So more selectivity is needed, an IF would have to account for the 32 kHz bandwidth. Is that correct, 5 Hz or DC to 32 kHz?. A massive segment of the 40 m band.

Cannot see how to use the unique signature frequency of the reference to discriminate the very narrow wanted part of all those 40 m signals!
I can get almost the same range by just amplifying the pick up across the room of the 32 kHz oscillator signal using a few inches of wire as 'antenna', with RF module disconnected.
The 32 kHz can be divided down, but then the advantage of keeping the reference away from the clutter of ordinary phone and CW would be lost.
Plus the advantage of the low pass filter of the lock-in amplifier reduces with relative frequency - the datasheet example is 400 Hz carrier with 0.1 Hz 'data' for 100dB discrimination through the noise.

Any thoughts?
If this could be got to work, in spite of twiddling a knob to adjust propagation phase shift, it opens a whole new field of QPR.

Thread Starter


Joined Mar 24, 2012
It is an experimental project.
Could you explain a bit further BR-549 please?

The premise is to improve the ability to detect very weak signals on receive.
Which would benefit QRP?

'inefficient way to transmit' could cover many things, perhaps you are looking at the use of a wide bandwidth to convey the same low data rate as simple CW?

Are you familiar with how a lock-in or phase detector works and can see some obvious hindrance to this ever functioning in the context of radio transmission?
Something similar has been used to recover transmissions from the space probes now located beyond our solar system.
Of course in those cases, the signal can be searched for with a sweep of frequency and phase shift, because the interference is noise. Not loads of other similar but much more powerful nearly adjacent signals.

Granted, so far I have not achieved anything more than can be done with a good pair of ears, and a very narrow bandpass software filter, set up on the precise offset frequency in use for test CW transmissions.
I could try dividing the clock reference with say a 4024 and NAND gate logic the outputs of a few stages, to create a more complex reference.
This would make alignment with a 'start sequence' an extra unknown sync issue to deal with.


Joined Sep 22, 2013
I hope you realize the difference between space communications and atmospheric communications.

And I still have no idea of what you are talking about.

What makes you think that an ultra wide AM signal is good for weak signal detection in our atmospheric noise?

Please explain your reasoning.

Thread Starter


Joined Mar 24, 2012
It is an ultra-narrow band AM signal. At morse 20 wpm 5 Hz frequency.

I have resolved some issues due to electrolytic capacitors causing small undulating DC shifts at the pre-amp audio stage.
I can now detect 32 kHz 5 Hz pulses using the pair of identical frequency crystal oscillators from hundreds of metres apart with no antenna at either end. I have had to hunt around the house locating every clock or device with a 32 kHz crystal and seal them in a screening metal tin.

Secondly, anyone would hazard a guess that the mention of space probe signal reception implies some basic understanding of RF reception.
But you are correct and welcome to throw in some good questions to consider carefully and get the concept grounded in knowledge rather than ignorant optimism.

In fact, you have prompted, perhaps unwittingly, a re-examination of the point at which envelope detection can take place. I did even try a straight AM broadcast detector diode arrangement at the RF stage as part of the experiments.
#And I can now see that it is at the low pass filter stage on the phase sensitive detector that the morse dits would be revealed.
Therefore, the required bandwidth is only a fraction of a Hertz.
The issue then is whether the ordinary radio CW traffic at 7.023 MHz will simply overload the RF input stage or IF stage.
I am designing a TRF stage with a 7.023 MHz crystal in the feedback (crystal tuned re-generative receiver always on the verge of oscillation) to experiment with how this might work to at least limit the amplified antenna signals to a very narrow frequency range.

Of course this may be a load of cobblers, like trying to substitute a piece of wet string(RF link) for a coax cable (true lab implementation of lock-in amplifier for nV signals buried in volts of white noise).
My practical understanding of PLL and lock-in has however vastly improved in this project.


Joined Sep 22, 2013
"I chose to use a 7.023 MHz carrier and AM modulated with a 32.768 kHz sine wave from a watch crystal oscillator, gated at 5 Hz (20wpm)."

"It is an ultra-narrow band AM signal. At morse 20 wpm 5 Hz frequency."

Sorry my friend, I am old school. I have a lot trouble understanding new concepts.

But one never knows. I have had some ideas about modulation and weak signal detection myself.

Thread Starter


Joined Mar 24, 2012
Old school is fine, now I understand where you are coming from, let me give a summary.

A lock-in or phase locked loop is a means to recover a signal from noise using the fact that only when the exact same frequency of a reference and the wanted signal match perfectly will there be a useful voltage output.

Normally, a direct copper wire links the reference to the electronic PLL or lock-in circuit (nowadays more likely an integrated circuit of many transistors in a black box chip).

What I propose is to replace the copper wire with a radio frequency carrier. 7.023 MHz is a convenient wavelength, my main antenna and activity is on this frequency.

Now, this 7.023 MHz is not stable enough without massive investment in ultra-stable oven-controlled crystal oscillators possibly slaved to an atomic clock standard in the shack. Unlikely that most hams will have an idle one lying around. A few Hertz wandering is no problem for rag-chewing, but hopeless for a reference for a lock-in amplifier. I can tell you most commercial rigs are wandering far more than their displays seem to re-assuringly indicate.

So what, I thought, if we add a sub-carrier to the RF, which is much easier to keep stable. And an easy to build and
I did some tests on 32.768 kHz crystals and confirmed they are ultra-stable. Not accurate, but stable.

The situation then is that there is an RF carrier at high frequency in the amateur band of 40m, with no severe demand on it wandering because the actual reference needed to weed out all the unwanted signals is modulated as amplitude variations. They do not shift frequency just because the carrier does so.

The sub-carrier of 32.768 kHz is now the reference for the exact synchronisation of the detection process. And it is known at both ends of the communication path, as if a physical wire connected the two, transmitter and receiver.

As mentioned previously, there are some caveats here to consider, such as knob-twiddling to accommodate propagation phase delays.
But if you have ever attempted to listen to CQ from the other side of the globe on contest weekends, there is a need for some imaginative new experiments.

Thread Starter


Joined Mar 24, 2012
And an easy to build, and resource components for, circuit. Ideally which can be an add-on to an existing transmitter and/or receiver.

My internet went down while I was completing the sentence above. Luckily the rest of the reply was saved in the cache.


Joined Sep 22, 2013
Let me see if I got this. Are you saying that if you know the exact transmit frequency, that we can detect that signal, even if the level of that signal is way below the noise level?

Or..... are you saying that to detect that signal, we not only have to know the frequency, but also the phase, to be able to detect it? And that is why you need a pilot sub-carrier?

"Normally, a direct copper wire links the reference to the electronic PLL or lock-in circuit (nowadays more likely an integrated circuit of many transistors in a black box chip).
What I propose is to replace the copper wire with a radio frequency carrier. 7.023 MHz is a convenient wavelength, my main antenna and activity is on this frequency."

What does that mean? In relation to this?

AND how do you handle that AM bandwidth at 7 MHz?

I must be missing something.

Thread Starter


Joined Mar 24, 2012
Yes, both the frequency and phase of the wanted signal must match exactly the reference.
Normally, this is achieved by a physical wire connecting the reference frequency oscillator inputs and the wanted signal modified by 'sensor signal chopping' (could be a wheel with slots in it, alternately obscuring a light beam and turning slow and minute changes in DC voltage into AC which is more easily amplified).
In the most sensitive lab experiments the properties of these connecting cables have to be accounted for with 'phase adjustment'.

What I am looking at doing is having a carrier No1 at RF, for the carrier No2 at 32 kHz. Not a pilot tone.
I thought I had understood that a CW signal has in theory zero bandwidth.
That is modified by having to switch on and off to transmit morse code. If a gaussian/sine wave shape is used, it sounds like a hum and is unreadable by humans. Hence, the bandwidth for morse is wider than the WPM would in theory imply. 20 WPM is about 5 Hz or 5 'dits' per second. It has more 'clicky' pulses which need about 100 Hz bandwidth.

Now I may have lost the thread here, but although the bandwidth appears to need to cater for 32 kHz, with a sine wave it is just a single isolated frequency sideband either side of the carrier. Conversely, the same situation prevails on receive. With straight AM there are two sidebands either side of the carrier at 32 kHz, yes. But nothing else between them and the carrier. I'm open to suggestion otherwise, because this is not something one can go online and just look up.
Indeed the space probe signals were recovered at their transmitted frequencies. Using rubidium state of the art frequency references to 'lock-in' to the original transmission frequencies a trillion miles away. With a whole team of expert knob-twiddlers. It was not the intended mode of reception, it is cumbersome and time-consuming to set up.


Joined Aug 27, 2009
Or..... are you saying that to detect that signal, we not only have to know the frequency, but also the phase, to be able to detect it? And that is why you need a pilot sub-carrier?
I think this is what he means.

1. You need a receiver capable of receiving the RF signal if the antenna noise floor was eliminated (~sensitivity/noise floor of the receiver) .
2. You need a system capable of detecting the RF signal in the presence of random antenna noise greater than the RF signal.

Generally you would use a spread-spectrum type system to do this. The end product for signal recovery is somewhat the same as over-sampling a ADC in the presence of Gaussian noise to get addition bits of resolution out of a slowly changing or steady source signal.

I'm not really sure what the OP is trying to do with his system of 5hz slices of the 32khz modulation on the RF carrier.
For a changing AC signal you need to be able to lock the detector phase (initial synchronization) with the expected signal somehow so the detector can average (with filter that might be digital) the antenna noise component to ~zero for a selected phase/time slice (5Hz data rate) of the total signal over many samples and/or use many slices of RF spectrum channels with random antenna noise on each to be averaged to ~zero while the same wanted deterministic signal is on each received channel.
Last edited:

Thread Starter


Joined Mar 24, 2012
Thanks for that link. I was aware of Dan Tayloes switched capacitor mixer, but it was useful to have another look at the ability it has to have selectivity at high Q within a small section of the 7 MHz band.
The design uses some high spec parts though. As a practical low noise mixer, it works but not as spectacular an improvement as the theory suggests.

WSPR mode uses a long period of sampling to retrieve weak signals, and of course there are many spread spectrum systems to defeat interference as well as improve signal to noise ratio.

I've no idea what I'm trying to do with a carrier and a sub-carrier to provide data at around 5 Hz beyond experimenting.
I haven't been able to do more than my ears could have picked out anyway. But then the signal at 5 Hz that I can retrieve from antenna noise is at a vastly better signal to noise ratio, it could be used to turn a tone on/off to replicate the original transmitted keying.
However, like I said in the first posts, a very narrow bandpass will do this just as well. With a nice 'drainpipe' acoustic.
I started playing around with this because I set up a software filter for a section of morse that I could hear marginally on a recording off-air, then realised that the same signal continued below the noise after fading. If I hadn't set up the narrow filter on an audible section first, I would never have known what to look for.
So I thought I could do better than the 'lucky dip' approach and have a pre-defined stable frequency to home in on.

Thread Starter


Joined Mar 24, 2012
I have had some success. Thanks for your comments and sorry to have had created some confusion.
I keyed the RF with a sine wave at 5 Hz. No transmitter 32 kHz modulation.
Rigged up a crude phase locked loop on the local oscillator, using the 32 kHz crystal still as a reference as well.
That only leaves the transmitter frequency drift.

As a Direct Conversion Receiver, the beat tone (at 32 kHz) is detected of many CW transmitters previously inaudible in the antenna noise.
As well as what I expect to see when I turn on my dummy transmitter at 5 Hz.
I can sometimes even 'read' the CQ characters in morse on the oscilloscope trace.

so this concept so far is little more than a demonstration that the 'straw' nature of antenna noise must have a large component of over-lapping transmissions. Interesting to see them on a trace displayed so clearly. Obviously I've no idea what the original noise ratio is, apart from being below 1, but the AD630 output shows the same noise has a signal buried in it which can now be measured with a 1 V peak envelope and only 2 mV peak noise.
Looking at some WSPR spectrum plots, there are a lot of low level normal speed genuine signals that must get 'filtered out' by the FFT sampling algorithm.
They show up in my tiny bandwidth (ie. signals that beat to 32 kHz exactly) even with propagation fade and transmitter drifts to contend with.

Thread Starter


Joined Mar 24, 2012
Thanks Dick for the interest. I put my camera batteries on to charge and was going to post some of my notes/schematics.
Because I am experimenting, there is not a 'system' in any shape or form, not sure how useful sketches would be.
It is just a 40 m wavelength loop in the attic, with part of loop as wide-spaced feeders taken outside and back downstairs to a current balun with soft iron core.
Small slug-tuned air-core transformer to a standard RF 18dB pre-amp and then a single balanced mixer with matched germanium point contact diodes.
AF first pre-amp is low input impedance based on audiophile power transistor design technique. About 0.8nV/Hz noise with gain of 100.
AD972 op-amp stages with flexibility for trying out filters, but for general listening just a gain of 8 each.
A computer usb powered speaker provides final audio. Total gain about 65,000.
Noise of antenna in a 3kHz bandwidth is about 30dB above all the input and mixer stages.

Researching QRSS, WSPR etc;-
I just wanted to try out the very different mode of oscillation and stability of a watch crystal as a 'crib', hence 'Enigma'.
Even keying the VK10M at 5 Hz in the drain of the tx output stage (causes no change in the 1 mA current draw) alters the frequency by a few Hertz. Whether by pulling the 7.023 MHz crystal via a minescule change in supply voltage or by changing the 'buffered' load capacitance, this is really sensitive compared to a tuning fork crystal run very gently in a low noise op-amp oscillator. It is simply impervious to these kind of effects, it takes a second or more to react to anything.
If all that has to be dealt with in a 'synchronous AM detection' is propagation phase shifts, this is much easier to cope with.

I have seen information on SDR using I and Q signals with a 'Costas' loop (whatever that is) in software to provide AM detection.
SDR is another route, I am playing around with the AD630 in the analogue realm, effectively doing the same averaging or integration of many short samples of the carrier frequency arriving in sync with the local reference.
To take advantage of the Direct Conversion Receiver with a single frequency I/Q process in an analogue stage would be handy, but I can't see how this would be done at the moment.


Joined Aug 21, 2008
Thank you for the overview. I was curious as to whether the integrator following the demodualtor in the AD630 is being used effectively. That's where a great deal of the SNR improvement in wider band (wider than QRSS) is obtained.

Thread Starter


Joined Mar 24, 2012
I didn't realise there were so many different modes active in amateur radio until I'd started this experiment and researching online.

One important factor with SDR is that it relies on the computer clock accuracy/stability of the host doing the sampling and FFT.
There are some expensive rigs available that probably have better stability than a USB dongle.

Thread Starter


Joined Mar 24, 2012

This link refers to the 1.46 Hz FFT bins for each tone difference in WSPR coding and the signal level having to be higher than the noise in that narrow frequency band.

A 'gain' in sensitivity of weak signal detection at -29dB is only referring to the original signal when stuck in the noise from a 2700 Hz bandwidth.
So it is not detecting 'below' the noise level at all.

If I can demonstrate detection of a signal which is too weak to display on Spectran for example (and WSPR signals are easily visible), and that signal is below the displayed noise in the 2.7 Hz FFT bins that Spectran is using?
The 'Transmitter' is just a well-shielded battery operated oscillator, kept a few metres clear of the antenna feeds and receiver.

Integrator bandwidth approximately 1/(2*Pi*time constant). So my experiment is 0.35 Hz, but the reference is 41000 times higher than this.
Does this make the difference? 32 kHz modulation would contravene licence conditions, but this is only a concept experiment.
WSPR is set up for centre of 1500 Hz beat oscillator, so relative frequencies to integrate are much less even though bit rate is longer.

How do you get a diode mixer to really misbehave? The need to 'avoid' AM modulation is often quoted.
For example, a diode forward voltage limits sensitivity as an AM demodulator. So if IF stage isn't suitable in this case (unless the mixer has the local 32 kHz reference also as the LO = 6.99 MHz IF or 7.055 MHz, nicely positioned IF ready to feedback/oscillate) can a diode mixer be skewed to use the reference at say 2f to forward bias. Leave one diode out?
I'm just curious as to how the same diodes allow detection of very weak signals just because they are in a mixer.

Thread Starter


Joined Mar 24, 2012
If someone can set up a simple beacon using a Pixie4 kit with the 7.023 MHz crystal? Leave out the trimmer fine tuning and all the rest of the current guzzling audio section. The basic oscillator should only consume 1 mA and will run fine on 5v (4 AA rechargeables).
Reduce the ERP drastically somehow, either by a resistor divider or source resistor, but useful if you can estimate how many mW it might be.
Anywhere in Europe more than 300 km from London , from 12 midday UTC (when the band is often down anyway) for 10 minutes or so every day.
Use a simple 1 Hz pulse square wave 'dit' key, or a single number in morse repeated if you have an arduino to pulse a VN10K or similar mosfet in series with the output transistor emitter (instead of the 'key').
I don't have to know what it is in advance, but I will recognise the signature from all the other weak morse I am picking up within a fraction of a Hertz.
Which signals to me on the oscilloscope look like random blips, sometimes a short sequence, while I'm trying to see how far down into the Gaussian noise I can go. I can't connect such a low frequency signal directly to my sound card to record, but I'll try and think of some way to get a hard copy.

Many thanks if someone gives this a go.
The chances of a random point to point skip, even with a good signal, are fairly low, so this would be a great go/no go check.