SAR image processing

Question

Is it possible for conventional x86 based processor/High precision GPU to be used for image processing for space borne Synthetic Aperture Radar (SAR) system? If not, are there alternatives using radiation hardened chips?

My primary concern is that the space grade RAD chips aren't powerful enough to process the heavy data output. Can a conventional processor be radiation hardened to be used for space applications?

I believe SpaceX uses redundant x86 chips http://space.stackexchange.com/a/9446/18573 — jayjay, Feb 20 '17 at 12:15
@Ken Adams Welcome to stack exchange (ok, three years ago!). You've asked three questions here and have quite a spread of answers. I suggest the part about "aren't powerful enough" could be refined so as to ask for the impacts of the SAR algorithm in terms of the scale of floating point operations required for a given image size and, if you got any answers to that, you'd have a good starting point for how long it would take on your candidate computer system. — Puffin, May 06 '20 at 12:49

score 11 · Answer 1 · answered Feb 20 '17 at 08:49

11

SAR (Synthetic Aperture RADAR) data is streamed to Earth-based stations for processing. x86 processors can and are used for deconvolving that mass of data. The processing load is high, but a modern desktop and GPU have the processing power of a supercomputer from 20 years ago. And it is easy on Earth to get rid of a couple hundred watts of heat in the electronics, which is a lot harder in space. Collecting and sending the data is much easier, uses fewer parts and less power, and increases reliability and lifetime of a satellite.

An Intel Xeon has a billion transistors. In rad-hard feature sizes, it would be a physically huge chip.

answered Feb 20 '17 at 08:49

C. Towne Springer

852
7
7

1

Synthetic aperture radar isn't always streamed to Earth. (At least) twice it returned to Earth on digital tapes. I'm guessing there are also cases when it is recorded and sent to earth at the next ground link opportunity (rather than continuous live-streaming). – uhoh Feb 20 '17 at 11:35
I think you're taking the word streamed too literally. It's sent to earth in raw form, because there's no reason to do the post-processing in orbit. – Tristan Feb 20 '17 at 14:57
@Tristan I think I am taking streamed appropriately literally. There is no internet ubiquity in space yet. Take a careful look at the infrastructure necessary when trying to keep a live data stream from the ISS to the Earth. Gbit/sec direct to ground needs a fancy slewing antenna. Until recently that would have been a dish, now it could be another phase array. Alternatively you can beam to another satellite M2M data infrastructure, if you can afford it. Everything is cheaper if you can do substantial buffering, which is not streaming. – uhoh Feb 20 '17 at 15:38
@Tristan Streaming is what is done with a live broadcast, and if you've watched live events from the ISS you know that those are carefully pre-arranged and time-limited because live-streamed are not easy nor cheap. Data would not be streamed. Yes, "sent" would be the right word, but "streamed" would not always be the right word. In space exploration it's good to take technical terms literally. This is not Quora. – uhoh Feb 20 '17 at 15:48
1

The lack of internet ubiquity in space is an appropriate caution against applying an internet-era interpretation of a term to a decidedly non-internet application. This isn't Quora, but being unnecessarily pedantic discourages newcomers (who may in fact be experts in the field) from becoming involved with this site. A precisely worded paucity of answers does nobody any good. – Tristan Feb 20 '17 at 16:20
@Tristan I'm sorry, but "streamed" strongly implies "sent in real time", rather than just "sent". This isn't an internet-era thing: the verb "stream" has meant "to be carried or emitted in a full and continuous current" since the 14th century (Oxford English Dictionary; subscription required). – David Richerby Feb 20 '17 at 18:27
Again, I ask, is this pedantry at all relevant to the question at hand? Whether or not something is streamed in real time is not at all germane to the question of whether SAR image processing is performed on orbit or in data centers on the ground. – Tristan Feb 20 '17 at 19:11
@Tristan I up voted and left a short, constructive, informative comment. Look at the comment again; "...not always streamed..." That's the nature of stackexchange at it's best. Also, in stackexchange answers are held to a higher standard than questions, and improving answers - even if by adjusting a verb - is part of what it's all about. It happens all the time here, and in other hard-science SE sites. – uhoh Feb 21 '17 at 00:20
4

I'm an old physicist. Here "streaming" is simply meant to mean "not processed in situ". Since the OP is about doing processing in orbit on Intel chips, I assumed that would be obvious without having to list all the options like taped, brought back digitally, brought back on a film reel and deconvolved optically, downlinks like SEASAT by microwave, anything but on board. Once down, it can be accessed on the line. – C. Towne Springer Feb 21 '17 at 06:01
2

I'm a systems engineer. Pedantry and "language lawyering" is what I do for a living. There's a time and a place for it; an informal q&a site generally isn't it, unless the misapplication of language seriously impairs understanding. In this particular case, insisting on a specific interpretation adds nothing of value to the answer, serves as a distraction, and presents a generally unfriendly atmosphere. @C.TowneSpringer points out that the word is only used colloquially here, which serves to answer the question as it is asked. We need not apply further judgment. – Tristan Feb 21 '17 at 15:06

score 6 · Answer 2 · edited Jun 17 '20 at 08:54

If for some reason you really wanted to do process Synthetic Aperture Radar (SAR) data on board a spacecraft, using commercial chips, you could do it. But it would be subject to several constraints and compromises.

Power dissipation:

Supplying kilowatt-ish electrical power would require large, expensive solar panels to keep pointed at the Sun, and a sophisticated cooling system to pull that same kilowatt out of the electronics as heat and then radiate it back into space. This could be a lot easier if you were operating at low duty cycle - if you could accept a lower duty cycle of measurements or only processing a subset of data to the highest spatial resolution.

Radiation hardness:

For non-mission-critical applications like this, you would not need the same level of rad-hardness as you would for spacecraft control. Since the calculation is so parallelizable, checkpointing and redundancy could make space-based computation viable with standard-issue Earthling CPUs.

(tl;dr if you make a mistake image processing, you can go back and do it again, but if you make a mistake and shut down an engine too soon, rebooting and trying again is a lot trickier.)

The mathematical process of analyzing the data is usually parallelizable, which means the problem can be broken into a large number of smaller problems, each solved separately, and then recombined at the end. A simple method to deal with radiation-induced numerical errors might be to simply solve every problem twice in two separate processors and when there are differences, throw them out and do it again. I'm sure there are more sophisticated ways as well.

Checkpointing is a systematic saving of intermediate results. Each time a small problem is solved you save it. Each time enough small ones have accumulated to calculate a larger intermediate result, you do it, and then save that. This way you can recover quickly from any radiation-induced crashes basically pick up where you left off.

note: I know your question is about x86 processors, the following diverges a little bit from those specifically, but it highlights the push to use more and more off-the-shelf computing power in fault-tolerant architecture to partially replace super-reliable rad-hardened but slow and expensive processors.

FPGA's are discussed within SXSE here: 1, 2, 3, in comments here; 4.

STP-H5 and SpaceCubes:

(more about SpaceCubes here and here.)

Part of the recent CRS-10 mission delivery to the ISS is STP-H5:

ISEM – SpaceCube, CHREC Processors

Validation of new processing technology is one of the most critical areas currently faced in the field of space technology given the rapid advancement of space-based instruments that require more and more processing power from highly reliable processor units. The STP-H5 ISEM (ISS SpaceCube Experiment Mini) will provide validation for hybrid-processing technology to advance that technology’s mission readiness level for future application on small satellites with limited volume for computer equipment.

In recent years, a shift has led from ‘radiation hardened systems’ to ‘radiation tolerant technology,’ that is technology which is susceptible to radiation-induced upsets but not radiation-induced destructive failures. Employing such systems can yield to a ten- to hundredfold increase in processing power at the cost of some science data due to occasional radiation related upsets that simply have to be tolerated. (emphasis added)

[...]

The v1.0 version has around ten times the processing power of the RAD750 OBC at half the power consumption; v2.0 has twenty times the processing power and requires 75% of RAD’s electrical power.

SpaceCube Mini was designed to similar specifications as v2.0 but in a 1U CubeSat form factor – fitting into a 10 x 10 x 10-centimeter volume. Its core is a multifaceted, hybrid space computer known as CSP, CHREC Space Processor. (emphasis added)

CSP combines fixed and reconfigurable CPU and FPGA processing functions using an algorithm that optimizes the use of the specialized resources of the device. A combination of commercial-off-the-shelf and radiation hardened systems creates a processing system with significant processor power and reliability in challenging space environments. (emphasis added)

above: from here, also described by Spacemicro.

Field Programmable Gate Arrays or FPGAs are finding their way into a wide variety of number crunching and signal processing applications. They can be competitive when the tasks have more diversity that would require very large or complex CPU architecture to provide high speed capacity, and especially if there may be a need to adjust or re-optimize signal processing architecture once the unit is in the field (or in orbit in this case).

A good example of that (use of FPGA for signal processing, not necessarily spaceflight) might be the FPGA implemented GPS radio "freely programmable front end and correlator with a custom firmware" mentioned by @Andreas in this excellent answer. The Piksi uses a Xilinx Zynq-7000 All Programmable SoC (AP SoC):

Zynq-7000 All Programmable SoC Product Advantages

The Zynq®-7000 All Programmable SoC (AP SoC) family integrates the software programmability of an ARM®-based processor with the hardware programmability of an FPGA, enabling key analytics and hardware acceleration while integrating CPU, DSP, ASSP, and mixed signal functionality on a single device. Consisting of single-core Zynq-7000S and dual-core Zynq-7000 devices, the Zynq-7000 family is the best price to performance-per-watt, fully scalable SoC platform for your unique application requirements.

Radiation levels:

Radiation in LEO is not nearly as bad as it is in higher orbits or deep space, except of course if there is some occasional unusual solar activity. As long as your application can remain in LEO and is not critical during solar events, this should all work.

Again, this is not necessarily competitive with doing the math on the surface, but for certain specific applications it may work.

score 2 · Answer 3 · answered Feb 20 '17 at 11:26

C. Towne Springer is correct; you can do a LOT more processing on the ground, and most systems do so.

That said, SAR processing is a multi-step endeavor, and it would certainly be possible to do some preprocessing onboard before download. The only savings, though, would be processing that results in a smaller, compressed data payload, since downloading is going to be far and away your greatest bottleneck.

SAR image processing

3 Answers3

Linked