Wednesday, September 22, 2010

setiQuest Deep Impact

The baudline signal analyzer was used to analyze the setiQuest Deep Impact satellite telemetry data file. Deep Impact was a 2005 NASA mission to crash a high velocity probe into the surface of comet 9P/Temple and make numerous measurements of the resulting explosion. That mission was successful and now the FlyBy section has been re-purposed to study other space phenomena. The Deep Impact satellite telemetry was captured by the ATA for this data file.

The quadrature data has a sample rate of 8738133.333 samples/second and since the meta data text file was missing the base frequency is unknown but believed to be in the X-band. The duration of the data file is 13 minutes and 23 seconds.

The following command line was used to stream the Deep Impact data files into baudline:

cat 2010-01-22-deep-impact-* | baudline -session setiquest -stdin -format s8 -channels 2 -quadrature -flipcomplex -samplerate 8738133 -fftsize 65536 -pause -utc 0

A Welch windowed 65536 point FFT for a 266.67 Hz/bin resolution was used to create the image below.

The main features are the suppressed carrier (+797199 Hz), the lower and upper sidebands (+597205 Hz and +997194 Hz @ +3.5 dB), and their harmonics (±400000 Hz). It is interesting that the separation between the center carrier and its primary sidebands is ±200000 Hz which is half the spacing of the harmonics. The rest of the band looks clean.

Note that the Histogram display has a nice Gaussian shape with even/odd holes because of the signed 8-bit sampling.

Zero Gaps
The Waveform window shows a gap of zeroes:

The zero gaps happen 3 times in this data file at these time locations with these durations:
  • 2:54.9, ∆0.177 s
  • 5:19.9, ∆0.115 s
  • 5:45.0, ∆0.099 s
The zero gaps are believed to be caused by errors in the data collection process. They likely are dropped UDP packets where the collecting process did not get a full buffers worth of data so it filled in the missing gap with zeroes.

+597205 Hz
Decimating by 512 for an effective rate of 17066.666 sample/sec and down mixing into the main lower sideband at +597205 Hz.

The Average window shows the center subcarrier with some tight modulation products that will be explored in detail later. The center subcarrier has ±800 Hz lower/upper sidebands and their harmonics have spacings of ±1600 Hz. This is the same frequency spacing scheme seen with the main +797199 Hz carrier but the frequencies have been scaled by a factor of 200000 Hz / 800 Hz = 250. [What is the significance of 250?]

Also of interest is the spectrogram display which shows that all the sidebands and harmonics are stationary in frequency at this zoom level. The Histogram has a small spike at the zero value that is explained by the zero gaps mentioned above. Other than that it has a nice Gaussian shape.

Here is the Average window zoomed in with the Hz axis set to 4X.

The three main tones are separated by ±800 Hz. The harmonics next to the center subcarrier have a spacing of ±39.098 Hz. The left/right sidebands have harmonics with ±39.098 Hz spacings. They also have a second set of ±39.098 Hz harmonics that have an offset of 18.262 Hz that likely originated from the center subcarrier (800 - 18.262 / 39.098) = 19.994 which is very close to 20. This means that harmonics of the center subcarrier are interspersed with harmonics of the lower/upper sidebands and they all share a spacing of ±39.098 Hz.

The spectrogram below is the same view as above but the frequency axis has been zoomed in with a Hz=1X setting. The subcarrier and harmonics all have a slight curvature.

Let us take a closer look at the strongest tone in the center by using a decimation by 4096 factor for a 2133.3333 sample/second rate.

This signal has a curved drift. The first half descends in frequency for a -3.06 Hz / 449 seconds = -0.00682 Hz/second drift rate. The second half ascends in frequency for a +2.08 Hz / 351 seconds = +0.00593 Hz/second drift rate. [Describe the possible motion that could cause this Doppler shape.]

+797200 Hz
Decimating by 512 for an effective rate of 17066.666 sample/sec and down mixing into the main suppressed carrier at +797200 Hz. Below is the Average display with a Hz=32X scaling.

The main center carrier has the same ±800 Hz lower/upper sidebands with ±1600 Hz harmonics that was seen above at +597205 Hz. What is missing are all the ±39.1 Hz harmonics. The main center carrier is much cleaner.

Below is the decimate by 4096 spectrogram view.

This spectrogram looks identical to the curved-drifting-random-walk seen above.

+997194 Hz
This is the upper sideband. The modulation and harmonic structures look identical to analysis that was done for +597205 Hz (lower sideband). The data for the spectrogram below was decimated by 4096.

This signal looks identical to the ones shown above at different frequencies. The same drift shape is not a surprise but I would of expected the content to be frequency inverted.

In this section the signal modulation of the +597205 Hz main tone of the lower sideband will be examined by down mixing and decimating by 524288. The effective sample rate is 33.3333 samples/sec.

Below is the Fourier spectrogram in magnitude space of an interesting section.

The middle section looks a lot like ten symbols of FSK with an alternating (01)* pattern and a delta between mark/space of 0.49 Hz. The periodicity bars measured a 1 /3.34 seconds = 0.299 baud rate. Below is the spectrogram of the same section of data but using the blip Fourier transform in phase space instead.

Directly below the FSK section mentioned above looks to be a region of periodic phase changes. Two distinct phases are visible which suggests BPSK modulation. Using the periodicity bars a 1 / 1.68 seconds = 0.595 baud rate was measured which is very close to double what was measured above.

Here is the blip Fourier phase spectrogram of a section a little farther down:

Take note of the numerous phase changes in the center section. Four distinct phases are visible which suggests a QPSK-like modulation scheme. Using the periodicity bars a 0.595 baud rate was measured again.

So we started with what looked like FSK and then jumped to BPSK and then to QPSK. Why? The clues are that the distance between FSK mark/space frequencies were very small and half the baud rate of the PSK measurements. Consecutive 00 and 11 symbols of BPSK can appear to look like very tight FSK. Then what appeared to be BPSK doubled its number of phases and became QPSK. Again QPSK can look like BPSK if only two symbols are being alternated. I believe the FSK / BPSK sections are actually QPSK of a sync header string such as (00001111)*(0011)* before the data sequence begins.

The only explanation I have for the extremely low baud rate is that the Deep Impact satellite is in a low power mode to conserve energy. A slower baud rate is also stronger and more tone-like which makes it easier for the Earth based telescopes to find it. This is only a guess since I wasn't able to find any information on-line about Deep Impacts modulation or telemetry schemes.

Self Similarity
Below is the Autocorrelation transform of the magnitude operation.

The horizontal and vertical features were most unexpected. The 3 vertical lines are periodic with a 572 sample (65.4 microsecond) separation. It was a difficult measurement to perform but the periodic bars proved to be very helpful in measuring a horizontal line periodicity of 1 / 0.01672 s = 59.82 Hz. So what is the significance of the 65.4 us and 59.82 Hz orthogonal lines?

Changing the point of view might help. Let us deconstruct the Autocorrelation transform down into its most basic primitive, the sample Raster transform. Below is a display of the sample Raster transform with its overlap width set to 572 samples in the Scroll Control window.

Three chirp-like whistlers are visible and they have a very LC discharge-like shape. There are thousands of these chirps and they are all slowly drifting to the left in time which signifies that the overlap width is fractional and not an exact integer. Assuming that this phenomena is in fact stationary a rough measure of the error is 13 / (145926.8 / 572) = 0.051 samples for a corrected overlap width of 571.949 samples (1 / 65.4469 us = 15279.6 Hz).

The periodic bars measured the chirp-to-chirp periodicity to be 0.01670 seconds which translates to a 59.88 Hz frequency which is suspiciously close to 60 Hz.

This temporally encoded chirp with a 15279.6 Hz line frequency and 59.88 Hz repeating periodicity is a surprising discovery that gets even stranger. Decimating by 2 and moving the down mixer frequency shows that this chirp signal is stationary. This is true all the way down to a decimation factor of 256 at which point the signal disappears. This means that this chirp signal passes through a tuner filter mostly unharmed. Note that this is the same "quadrature magnitude elephant" phenomena that was first seen in the Exoplanet 060 analysis report but the quadrature magnitude operation isn't required to see this particular LC chirp. Another difference is that the Exo 060 sighting focused on 1/3 Fs harmonics and different 25600 Hz subharmonics but 60 Hz was present there too.

Something interesting happens as you continue to decimate down while tuning into one of the sidebands. Here is the Autocorrelation transform when decimating by 4096 and down mixed to the +997 kHz upper side band.

This again was not what was expected. The peaks have a periodic spacing of 54.611 samples which works out to a repetition frequency of 1 / 25.534 ms = 39.064 Hz which was also seen above in the sideband harmonics. Changing our point of view helps show us what is going on. Here is a spectrogram of the sample Raster transform applied to the magnitude operation with an overlap width of 764 samples. Click on the image to see the fine details.

The 14 vertical bars have a thickness of 8 samples (3.750 ms) and a very slight drift to the right with a slope of 9 / (612267 / 764) = 0.011 samples. This is the error term for a corrected overlap width of 764.011 samples (358.1 ms). The 3 longer horizontal black lines have time stamps that correspond to the 3 zero gaps mentioned above. Also of interest are the 40+ shorter horizontal lines that are bounded by the vertical bars. They are not zero gaps but have decreased amplitude. Their origin is unknown and their spacing appears random.

This vertical bar effect is not a function of the magnitude operation. Here is a decimation by 2048 view with the green and purple colors representing the I & Q channels. The changing wavy pattern is the main sideband frequency that is slowly drifting. So the vertical bars are stationary while the sideband signals drift.

Why the vertical bars? Is the 25.534 ms spacing significant? What do the 40+ bar-to-bar amplitude dropouts represent? I don't know these answers but some very complicated signal artifact-ing is taking place.

Filter ExtractionThe impulse response transform was used to compare the I and Q channels. To inverse the positive lag symmetry a Hilbert filter was applied to the I channel prior to calculating the impulse response. This effectively undoes quadrature. Note that the horizontal axis should be time lag (not Hz) and the vertical axis should be a linear scale (not dB).

Compared to all the other setiQuest Filter Extractions I've done the filter shape and the sharp notch at zero are upside down! Another unusual feature is the strong periodic ripple throughout the whole filter. This file had strong carrier tones which is not a good stimulus source like noise is. So this strange periodic ripple could be due to the carrier tones or to a very broken quadrature filter which is possible since this data file is from January 2010.

The Deep Impact signal has a very rich self-similar harmonic structure at multiple levels of zoom. At the top level a slightly suppressed carrier has sidebands that have a spacing of ±200000 Hz with weaker ±400000 Hz harmonics. Zooming in on one of the sidebands reveals a sub-sideband spacing of ±800 Hz with weaker ±1600 Hz harmonics. What is the significance of 200000 Hz / 800 Hz = 250 ratio of the harmonics? Zooming in further shows 39.1 Hz harmonics. Is the rich harmonic structure due to Deep Impact having a non-linear amplifier for radio transmission which is common in deep space probes for power efficiency reasons or are the rich harmonics an ATA collection artifact?

Zooming into any of the strong sidebands or harmonics reveals a drifting signal that looks like a mixed FSK / PSK modulation scheme at a low 0.595 baud rate. I believe the FSK section is actually QPSK of a repetitive header sequence. Deep Impact had a downlink speed in the Mbps range so why do I measure such a very low baud rate? Is Deep Impact in a low power sleep mode?

The Autocorrelation transform found something unusual and very unexpected. Further investigation with the Sample Raster transform shows abundant whistler-like chirps with a ~60 Hz periodicity. Zooming in with decimation by 4096 shows an Autocorrelation periodicity of 39.1 Hz that the Sample Raster shows to be vertical bars with encapsulated dropouts. Note that the finest harmonics of the upper/lower sidebands had the same ∆39.1 Hz spacing. What is the significance of the bars, the dropouts, and the relationship to the sideband harmonics?

This Deep Impact data is an excellent example of what a SETI signal might look like. It is unknown which analysis components are caused by the satellite and which are artifacts of the collection process. This is an important distinction but it really doesn't matter for the purposes of this analysis report because the rich harmonic structure, modulation, and raster features all have an equally fascinating complexity.

Data licensed through SETI.
Software licensed through SigBlips.

No comments: