MH370 and WSPR: Aircraft Scatter on HF – A Critical Review

Some articles had been published stating that processing of WSPR logs can assist in reconstructing the route of an aircraft (i.e., MH370) by a method known as bistatic radar. Nils Schiffhauer, DK8OK, has some doubts. Read his reasoning below.

Some Doppler traces in vicinity of the carrier of China Radio International from Xian-Xianyang, broadcasting with 500kW (250kW carrier) on a curtain array antenna providing nearly 25dBi gain towards 190°. More on the lower than on the upper sideband you see some aircraft Doppler traces at distances of 8Hz to 17Hz (LSB), corresponding to a relative velocity of the aircraft from about 200 to 600 km/h. Frequency resolution: 0.0047 Hz.

Again and again, efforts by radio amateurs make the rounds to have identified the crash site of MH370, for example, on the basis of the evaluation of WSPR logs. For this purpose, they primarily evaluate “unusual” level changes in the WSPR logs.
The fact that RF signals are scattered by the metal hull of aircraft is nothing new. The best way to see this effect is to look at the Doppler tracks that form at a certain distance from the actual carrier frequency of the transmitter. This is based on the theory of bistatic radar (see for example “Bistatic Radar” by Nicholas J. Willis, Raleigh NC, 1995).

Aircraft Scatter: Bistatic Radar

This concept has been used worldwide for decades, in the HF range primarily in the form of various over-the-horizon (OTH) radars.
In principle, a signal with known properties (amplitude, waveform …) is transmitted and received again after having passed through the “channel”. Comparing the transmitted signal with the received signal, the properties of the channel can be deduced.
Professional systems with powerful transmitters, beam antennas with high gain and signals with precise characteristics, whose scattering is evaluated with highly specialized algorithms, allow the detection of even small aircraft and ships in rolling seas. Intelligent evaluation includes the extensive elimination of interference factors, from sea clutter to changes in the ionosphere. These also have a considerable influence on the received signal – including amplitude and frequency (Doppler and split-up into x and o rays).

WSPR: A challenging mode

Transferring this technology to the evaluation of level and frequency of WSPR signals faces a couple of challenges:

  • The effective transmit power of WSPR signals is only roughly known at best, but in terms of magnitude it is at least 50 dB below that of professional OTH equipment.
  • The quality of the transmitted signal in terms of frequency stability, noise and possible amplitude changes (power supply!) is not known.
  • The changes of the ionosphere (attenuation, Doppler, multipath …) is not known.
  • The waveform WSPR, as well suited as it is for QRP communications, has not been developed for its use as bistatic radar.

All previous evaluations, especially in connection with flight MH370, are based primarily on evaluations of level changes, measured as just one mean value of a 110 seconds long transmission. It has been postulated that aircraft scatter increases the overall level of the signal. (“Drift” seems not to be a proxy for “Doppler”, see below). A possible evaluation of the Doppler shift fails so far largely because of the data situation. This also prevents the inclusion of the current state of the ionosphere and its local fine structure. Catching up ray tracing – moreover only two-dimensional! – can by no means compensate this disadvantage.

Correlation vs. Causality

However, far-reaching expectations are attached to the existing and modest material [Sensational new finding for MH370 flightpath], which in my opinion are already epistemologically, but also technically on feet of clay. This is like one can make Mozart’s “Kleine Nachtmusik” sound out of pure noise by suitable filtering. This also happens to high-ranking scientists, for example, when they detect neuronal correlates in the brain of a dead salmon and erroneously conclude that it is solving mathematical problems … there has been everything. Correlation doesn’t always mean causality.

My doubts about using WSPR logs for the purpose of locating aircraft are based on two main points:

  • We know little to nothing about the actual state of the fine structure of the ionosphere, which primarily affects the signal.
  • We have only guesses about the extent to which (amplitude, Doppler …) an object flying in an unknown direction scatters the signal (type, height, direction, speed …).

Furthermore, raytracing in those examples calculates with unusually low elevation angle of under 3°. PropLab 3.1 [most recent build #43] sees the main elevation of, e.g., a vertical dipole at more than 20°.

Just as an aside, when the central paper on “WSPRnet Propagation Technical Analysis” states, “Flat ice or calm ocean provide the best surface’s for WSPRnet signal reflection.”, the opposite is true, at least as far as the conductivity of ice is concerned. It has the worst conductivity and thus reflectivity of all soils on the earth’s surface and, at 10-4 S/m, is in last place in this respect according to ITU-R P.527-4.

HF Scatter: What the Experts say

The relevant literature on HF radar deals with such central matters as radar cross section (which defines the return power of an modeled object, which can vary by several 10 dB under different circumstances), scatter in Rayleigh and Mie regions (dependence on wavelengths and dimensions of the object), and inhomogeneities of different layers of the ionosphere – concepts of which most previous studies on the subject of “WSPR and MH370” make sparse use at best.
I do not want to bore now with long-tongued recounting of these things, but to pick out thesis-like only some points from the NATO paper HF-OTH Skywave Radar for Missile Detection” as a quotation (bolded by me):

* We must deal with heavy propagation losses due to the very long travelling distances as well as strong absorption losses mainly due to the D layer of the ionosphere. The whole loss contribution can be up to 100-150 dB.

* The apparently simple propagation mechanism hides the complexity of the ionosphere structure. This implies a challenging target localization that could be achieved by a smart system calibration combined with a three dimensional reconstruction of the signal path through the ionosphere.

* OTH radar system functionalities are strongly dependent on the ionosphere and on the environment noise level that means geographically dependent performances. Accordingly the radar siting represents one of the key choices.

* High values of peak power are necessary in such systems to deal with strong losses.

* It is essential a simulative approach that can provide a predicted radar cross section variability as a function of the operating frequency and of the aspect angles that are unusual for ordinary radar systems.

Certainly, these military requirements do not have to be transferred 1:1 to our more semi-professional approach. Nevertheless, they set such narrow limits to even our modest approach that their meaningful application threatens to disappear almost completely in fading and noise.

Amateur’s Choice: +70 dB – and more

In order to verify at least some basic assumptions in practice, I have conducted a series of investigations on carriers of shortwave broadcast transmitters. These have several key advantages over WSPR:

  • They provide a known signal in terms of frequency stability and amplitude.
  • Their effective transmit power is about 70 dB higher than that of WSPR transmitters.
  • Each station transmits continuously for at least 30 minutes, allowing relatively large integration times to improve the frequency resolution (up to 0.005 Hz, or adding another 57 dB in a 2’500 Hz channel) and thus the sensitivity for detecting the Doppler signals.
  • This setup allows a clear separation of original signal (via ionosphere) and aircraft scatter by shape and frequency. This eliminates what I consider to be the biggest unknown of the WSPR approach.

However, there is one drawback: only between 0 and 10% of even these strong transmitters can be used to detect Doppler traces from aircraft. Only with a smart match of frequency, time, flightpath, propagation … you will have some success.

Let me pick a typical example from a series of experiments – all with similar outcomes. The Chinese transmitter at Xian is received with 500 kW (carrier: 250 kW) on a GPS-locked SDR Elad FDM-S3. Its carrier on 17’530 kHz was first displayed under magnifier of two software, namely PSKOV (screenshot on top of this blog, for a first introduction click here) with 0.005Hz resolution, and SDR software SDRC with 0.39 Hz resolution, see screenshot below. The latter provides the numerical output used in the following post-processing.

Situation with software SDRC, compared to the PSKOV-screenshot at top of this blog.

Needle in a Haystack?

The next step is to answer the following question: Is it possible to detect the influence of the Doppler tracks in the overall signal? After all, this is the method that is tried using WSPR.
For this I first divide the total signal of 100 Hz width into three channels: Carrier, Doppler and Noise, see below. The Doppler signals are clearly visible, the respective correlation coefficients between the channels “Carrier”, “Doppler” and “Noise” are all well below 0.1, which shows the independence of the three different signals from each other and their separation. Mean Level of Carrier is -59.1 dBm, standard deviation 5.71. Mean Level of Noise is -108 dBm, standard deviation 4.5.

Signal levels over 10’000 seconds for all three channels.

In the following step, I sum up the channels “Carrier” and “Doppler” (de-logarithmize the dBm values in mW, addition, re-logarithmize the mW values to dBm).

If I now compare the data “Sum level of Carrier and Doppler” with the “Carrier”, the correlation diagram shows a near-complete agreement between both data sets – see screenshot below:

Correlation diagram fo Carrier and Carrier+Doppler (blue marks) shows only miniscule differences from 100% identity (red line).

Are rounding errors the reason for those miniscule differences? No, as we see in the following screenshot where you can see that the Doppler trace increases the signal of the carrier by 0.2 dB to 0.3 dB at most, except for a single exception: 0.6 dB.

But these values are almost completely lost in the overall signal with its standard deviation of 5.71, if they are not anyway below the measurement accuracy of many SDRs and their software.

The screenshot below draws the difference between [Carrier+Doppler] and [Doppler] – values left Y-scale – together with the original Doppler signal – right Y-scale. Judge for yourself …

The Doppler traces are practically lost in the fluctuations of the carrier signal.

For the complete overwiew of the steps see the workflow below:

Workflow, done with Elad FDM-S3, SDRC, MatLab R2021a.

WSPR & Aircraft Scatter? I have my doubts.

We started with the challenge to see some signal enhancement by scattering from aircraft in WSPR logs. Those should lead to “unusual” changes within the signal. A WSPR transmission lasts for 110.6 seconds and delivers just one mean SNR value representing his time (plus drift with a resolution of 1 Hz only).

It was suggested that from these signal levels (and drift data) aircraft scatter can be derived. This had been tried to underpin with 2-dimensional ray-tracing propagation simulation, based on statistical, rather then real data.

I tested those assumption with 10’000 level data at one second resolution, +70 dB in transmitting power, added by a few 10 dB of processing power. Doppler traces from 30+ aircraft had been analyzed. Backed by this, it can be stated:

  • On HF over longer paths (from two hops/with multipath propagation), usual aircraft scatter has nearly no effect on the overall reception level. Without prior knowledge, it is hard, or even impossible, to conclude aircraft scatter from the sum signal.
  • Doppler effects occure in the region of about 5 Hz to 20 Hz and don’t coincide with the much lower “drift”, I saw in the WSPR logs.
  • The power of a typical WSPR setup is many ten dBs down to what it should be to reliably identify aircraft scatter.
  • We usually know near nothing about the transmitter’s and receiver’s site – power, noise, drift etc.
  • We know near nothing about the channel (propagation) at the refracting points. This makes it difficult to separate different effects from each other, of which aircraft scatter is just a minor one, with multipath fading being the absolutely prevailing one.
  • The statistical population/resolution of the data one gets from the WSPR logs is too small (due to the 110 seconds) to apply robust statistical methods to cope for a dynamic environment.
  • The simulation capabilities of PropLab are not sufficient for such long-range statements due to, for the given case, poor temporal and spatial resolution of the usable ionospheric data. In addition, the simulation with PropLab was, to put it mildly, not optimally implemented – 2D instead of 3D, airy assumptions about the angle of incidence of the signal, and wrong assumption about ground conductivity.

My final conviction is: the detection of aircraft scatter and its assignment to specific flights from WSPR data is far more wishful thinking than reality. Only with considerable prior knowledge and using other data sources as well as possible coincidences these statements can be explained. WSPR in itself is not likely to contribute to this.
There are far more RF signals that are far better suited for this purpose for a variety of reasons. Trump, however, would be digital beacon project, whose waveform is suitable for also qualitative studies of the propagation path. Here, a private initiative seems to be active, after amateur radio clubs continue to stick to analogue technology (NCDXF, once meritorious).

Just a quick answer …

… to a question, I have been asked:

Q: “WSPR is sampling SNR for 110 seconds, boiling this down to one value. The resolution of your approach is one second. Does this influence the results?”

A: “Surely – higher resolution = more details, better insights!”

Q: “Can you show me?”

A: “I warn curious, but that you are, gladly!”

The screenshot below offers an answer to the question: “What does the WSPR log see, which only notes the sum voltage and also this only as an average value (?) – and also this not every second, but in intervals of 110.6 seconds?” Because with WSPR one has only this one sum value. For this purpose, exactly this situation was simulated with the originally good 10,000 data values collected every second and these were divided into groups of 110 seconds each, whose mean value was formed. This then corresponds to the SNR value in the WSPR logs.I boiled down my data of “Carrier+Doppler” and “Doppler” to 91 groups of 110 seconds each, and then I calculated the mean values of each group. In my view that should match the SNR values of WSPR (“Carrier+Doppler”).

Sum level and Doppler level. Allegedly, one should be able to conclude from “unusual” sum levels to aircraft Doppler. Which without prior knowledge leads predominantly to false-positive (FP) and false-negative (FN) results, of which only some are entered here. Rolling the dice gives a better result.

I then calculated the 95% confidence interval from these largely normally distributed mean values and restricted the plot in the above screenshot to those values that lie above the upper limit of this confidence interval of 58.02 dBm. These are “unusual” values in my eyes, although this is only a reasonable guess, because the authors do not specify further what they mean by “unusual”. The course of the values “Carrier+Doppler” is scaled on the left Y-axis, from -58 dBm to -54 dBm.
According to theory, a Doppler signal should now be lurking behind each of these values above -58 dBm, at least in the “unusual” peaks. Is this true? To check this, I plotted the corresponding processed Doppler levels in the same diagram, scaling on the right Y-axis, -110 dBm to -98 dBm.
Already a first look shows that the connection of both curves is rather random. With FP like “false positive” I marked for clarification at least some of those positions, where due to the “unusual” sum signal an aircraft Doppler would have been expected – but was not present. With FN are marked, vice versa, some of the positions, where there was a Doppler signal, but no “unusual” course of the sum level indicated it. A clear assignment “unusual sum signal -> aircraft Doppler” is therefore unlikely.

Comments welcome: dk8ok at gmx.net

P.S.

Oh, yeah, we had some comment! Richard Godfrey rebuked my technical-scientific based criticism on his website (“Serving the MH370 Global community”), which explicitly invites discussion, with the following “arguments”:

“You are not welcome on this blog!You were thrown out of the German Radio Amateur Club (DARC) in 1992. Despite 3 appeals at regional and national level as well as in court, you are still excluded from membership 29 years later. There are very good reasons for this. …I have complained to Christian Entsfellner DL3MBG, the current Chairman of DARC, about your demands to protest against my work and that of Dr. Robert Westphal (DJ4FF) officially on the DARC website regarding MH370 and WSPRnet.Your paper is plain wrong and your arguments are misplaced.I suggest you go elsewhere as I am sure there are other MH370 websites who will welcome the likes of you. Und Tschüss!”

” Ambition should be made from sterner stuff.” [Shakespeare, Julius Caesar, Act III]

I like to recommend a website where you find much ado about sterner stuff concerning MH370 and WSPR:
MH370 and Other Investigations – Following the Data Towards Discovery.

3 comments

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s