Sunday, March 22, 2020

aprs.fi supports Kenneth's Proposed Telemetry Format

Telemetry is the collection of measurements or other data at remote or inaccessible points and their automatic transmission to receiving equipment for monitoring. The word is derived from the Greek roots tele, "remote", and metron, "measure". (Wikipedia)
APRS telemetry can be used to transmit things like voltages, temperatures, and channel traffic status from anywhere to everyone on the APRS network.

Today I've changed how aprs.fi deals with telemetry values. The change is backwards compatible, but if you start to play with this on the transmitting side, strict classic implementations may not accept your data any more. Please read the whole article if you're going to make use of this.

The APRS protocol is documented in APRS101.PDF. Chapter 13 specifies how telemetry can be transmitted. Analog values are defined to be 8-bit integers:
There are five 8-bit unsigned analog data values (expressed as 3-digit decimal numbers in the range 000–255), followed by a single 8-bit digital data value (expressed as 8 bytes, each containing 1 or 0).
In year 2020 you may find this a little bit restrictive, as a lot of measurements are done at slightly higher resolution these days. Even the popular DS18B20 temperature sensor does 12 bits. Oh, and it can do negative temperatures too.

Values that are negative, or larger than 255, or decimal values which fit within 8 bits, are supposed to be scaled to fit within the 0...255 range. A separate EQNS Equation Coefficients packet should be sent to provide coefficients a, b and c, which are used to scale the transmitted value v back to the original measured value X using the formula:
X = a * v2 + b * v + c
So, you could scale temperatures between -40 and +60 Celsius to a range of 0..100 by adding 40, and then transmit an EQNS packet saying a = 0, b = 1, c = -40, and the recipient could then figure that a transmitted value of 60 would be 0*60^2 + 1*60 + -40 = 20°C.

But if you follow the spec, there's no way to transmit a 80-degree temperature range (-40°C to +40°C) with a 0.1°C resolution. It could be scaled to 0..800 range, but it'd overflow the 8-bit upper limit of 255.

One of my in-house temperature & humidity sensors.
ESP8266 on wifi + DHT22 compatible AM2302 sensor.
Not APRS though, MQTT to Graphite & Grafana.
Needless to say, a lot of people found this unnecessarily complicated and limited, and figured they could just ignore the slightly archaic specification and simply transmit and accept values larger than 255, decimal values, and negative values. When I originally implemented telemetry decoding for aprs.fi in 2008, I already found a lot of this happening.

I figured there must be other implementations that are more relaxed than APRS101.PDF and just went along, picking a backwards compatible but slightly more relaxed schema using the Stetson-Harrison method, without really documenting the situation. So, until today, aprs.fi has been storing an unsigned fixed-point decimal number which can represent values between 0 and 999999,99. Two decimals. No negative values. It has been stored as a 32-bit data unit in the database. I admit it's been a bit silly - I didn't go all in and accept negative values, and the fixed point is a bit... fixed.

Kenneth Finnegan (MS in APRS, the highest academic degree anyone has on the subject) has written a some good notes on how APRS telemetry is actually formatted in the wild here:

https://github.com/PhirePhly/aprs_notes/blob/master/telemetry_format.md

Kenneth also outlines a Proposed Telemetry Format, where a telemetry packet can be made shorter if there's less than 5 values to be transmitted, and where large, negative or floating-point values can be transmitted as-is without scaling them to the classic nintendo 8-bit 0...255 range specified in APRS101.PDF. I think the proposal is good, and it also matches the reality of what is already being done widely by many telemetry transmitters.

The proposed format is backwards compatible in a way that allows decoders accepting the new format to also accept the old format without special handling. Old and pedantic decoders can not accept the new format when either the shorter format (with less values) or decimal/negative/large values are sent. Luckily it's easy to upgrade the old decoders to accept the new format, and even the smallest microcontrollers today are easily programmed to handle floating point values like this.

As of today, aprs.fi implements Kenneth's Proposed Telemetry Format.


  • Analog values are stored in signed IEEE single precision (32-bit) base-2 floating point format. Kenneth does not specify a resolution or maximum value, so I just picked the single precision format, since it doesn't actually increase my disk storage requirements. As the wikipedia article says, the single precision float can store a maximum value of 3.4028235 × 1038. "All integers with 7 or fewer decimal digits, and any 2n for a whole number −149 ≤ n ≤ 127, can be converted exactly into an IEEE 754 single-precision floating-point value."
  • The decoder only accepts values between -2147483648 and 2147483647.
  • Since it is a floating point value, if you send very large values, the resolution starts to show up in the less significant digits. But it is still exactly precise in the -9999999 to 9999999 range, and certainly in the 0...255 range too.
  • Only three decimals are printed on the telemetry pages for now - if you transmit 0.0005 it'll be rounded up to 0.001. This is because the values calculated with EQNS coefficients often have a lot of decimals, and for most data it does not make sense to print it out with more than 1, 2 or 3 decimals. But I'm guessing there isn't much data that would have meaningful 4 decimals.

A few examples of packets that can now be accepted

Short telemetry, just one value of 42:
T#001,42
Slightly longer, with three values, first one being negative, second one being large, third one having decimals:
T#042,-1,10000000,142.4242424
Just four bits of digital telemetry:
T#999,,,,,,1101

Caveats

Older decoders which read APRS101.PDF to the letter will not accept these packets. I'm hoping they will be relaxed to accept the proposed format in the future. At this point you'll need to test if your intended recipients parse them.

javAPRSSrvr will pass these packets through, but if a client sets up a "t/t" filter to request all packets of the Telemetry type, it will currently only accept packets which strictly match the APRS101.PDF format. If there is a decimal point in a value, a negative sign, or an analog integer value exceeding 255, the client will not get the packet. I'm hoping Pete will relax this requirement in the future.

aprsc packet type filters (like "t/t") only look at the packet type (first few bytes usually). It does not care if the remaining packet matches some format or not, and it allows upgrading the format without changes in server code.

Final note to people writing APRS decoders

If you implement the Proposed Format, please let me or Kenneth know, or make a github pull request to Kenneth's document, so that we can have a list of software supporting the new format. Thanks!