January 2013: The long lost documentation
January 2014: Updates for rtl_fm overhaul

Original The long lost documentation Updates for rtl_fm overhaul Editable
version 1 & 2 of 2


Rtl_fm is a little utility I wrote for the rtl-sdr project. The program was made to fill a gap in software defined radio: all the computers weaker than a Pentium 4. Basically, an Atom processor processor does not have enough oomph to demodulate something as simple as narrow band FM using the standard tools. (Recently a high performance FM demodulator was released, Simple FM but it works only passably on newer Atoms.) So rtl_fm was written with one goal, efficiency, in mind.

Greg (2013-05-05-15-14-58-896)

It also works nicely on a Mac under OS X assuming you have all the libraries it needs.


The very first version ran with plenty of cycles to space. As an unexpected bonus, it ended up being efficient enough to easily run on small ARM boards such as the Raspberry Pi. GnuRadio is a really great program easily worth a thousand bucks. But it was designed to run on thousand dollar $500 computers with thousand dollar $500 SDR hardware. Where as this is made for $20 SDRs plugged into $20 computers. More powerful windows and OSX machines can easily handle several instances of rtl_fm at once


Rtl_fm is a general purpose analog demodulator. It can handle FM, AM and SSB. It can scan more than a hundred frequencies a second. For digital modes, piping the audio into multimon-ng works very well. On a Raspberry Pi, rtl_fm + multimon uses just over half of the CPU. If you are interested in ADS-B decoding, check out the rtl_adsb utility. It uses only 6% CPU on a Raspberry Pi and can run on targets as small and adorable as a 16MB TL-R703N. If you are using Arch Linux, I've already packaged these tools for you and you can install them with pacman -S rtl-sdr. If you need something more specific, feel free to send me an email.

kk (2013-10-11-09-38-51-405)


Throughout this guide, I'll be using Sox's play command to play the audio. This is because Sox works on Linux, OSX, Windows and pretty much anything else you can run a compiler on. If you are using Windows, rename the sox binary to play first. (If you are using Windows, rename the sox binary to play first.) Sox will automatically resample the audio to keep your soundcard happy and can apply denoising filters to keep your ears happy. If you want to run closer to the metal, feel free to use the utilities provided by Alsa or OSS or Pulse. But I won't be talking about them in this guide.

Before starting, make sure your dongle works.
rtl_test should return a list of supported gain values and not produce error messages. Make sure you are using a USB 2.0 port as well, otherwise you get really weird errors.

It is also very important to always make sure rtl_fm and Sox are both set to use the same data rate. This is probably the most common problem people run into.

Music on the Radio

rtl_fm <span class="del">-W</span> <span class="add">-M wbfm</span> -f 89.1M | play -r 32k -t raw -e <span class="del">signed-integer</span> <span class="add">s</span> -b 16 -c 1 -V1 -


This will tune to a broadcast FM station.
-f gives the frequency to tune in. -W -M wbfm says to use wideband FM mode, but this is really a shortcut for a tweaked narrowband FM mode. It expands fully into

kk (2013-01-30-12-39-19-716)

Add a note that if it sounds bad, the antenna is bad and the gain (-g) should be set manually.

kk (2013-01-30-12-43-07-997)

... And to get a list of legal gain values with rtl_test.



rtl_fm -f 89.1M <span class="del">-N</span> <span class="add">-M fm</span> -s 170k <span class="del">-o 4</span> -A <span class="add">fast</span> -r 32k -l 0 <span class="del">-D</span> <span class="add">-E deemp</span> | play -r 32k . . .
kk (2013-08-17-22-04-04-901)

-A fast


  • '-f ...' indicated the frequency to tune to
  • -N -M fm means narrowband FM
  • -s 170k means to sample the radio at 170k/sec * -o 4 indicated to sample at 4x the requested rate and downsample later
  • -A -A fast uses a fast polynominal approximation of arctangent
  • -r 32k means to lowpass/resample at 32kHz
  • -l 0 disables squelch
  • -D -E deemp applies a deemphesis filter

Needless to say, this is a lot just to listen to the local radio. So use -W -M wbfm instead. If conditions are poor you might need to manually add some gain with -g 20 or so.

Police Scanner

Operating police scanners is illegal in some countries, notably Germany.

rtl_fm <span class="del">-N -E</span> <span class="add">-M fm</span> -f 154.42M -f 154.75M -f 154.82M -f 154.89M -s 12k <span class="del">-o 4</span> -g <span class="del">49.2</span> <span class="add">50</span> -l 70 | play -r 12k . . .

This one will take a lot more fiddling to work properly. First, you of course need channels used by your municipality. And this will only work if your local services have not upgraded to P25 or some other trunked system.

  • -N -M fm narrowband FM * -E edge tuning. This gets a whole paragraph of explanation below.
  • -f ... frequency to tune, use multiple times for scanning
  • -s 12k sample rate, about as narrow as possible for FM voice * -o 4 same as above
  • `-g 49.2</span> <span class="add">50 set gain to maximum (use a value appropriate to your dongle)
  • -l 70 set squelch to 70. The exact values varies a lot. Changing the gain or sample rate will require a change in the squelch to compensate.

Airband Scanner

The airband has a lot of channels to scan. Explicitly listing them all is not practical, so use a channel range.

rtl_fm -M <span class="add">am</span> -f 118M:137M:25k -s 12k -g <span class="del">49.2</span> <span class="add">50</span> -l 280 | play -r 12k . . .
  • -M -M am AM demodulation
  • -f start:stop:interval a range of frequencies to scan
  • -s 12k same as above
  • `-g 49.2</span> <span class="add">50 same as above
  • -l 280 squelch level, exact value varies a lot

Pager Decoder

*Pagers are considered private communications in some areas (notably the US) and may be illegal to read.*

rtl_fm <span class="del">-N -E</span> <span class="add">-M fm</span> -f 929.77M -s <span class="del">22.5k -o 4</span> <span class="add">22050</span> -g <span class="del">11.5</span> <span class="add">10</span> -l 250 | multimon -t raw /dev/stdin

Exact frequency and gain depends on your area. Most of these options should be familiar by now. The only new thing is `-s 22.5k</span> <span class="add">22050 (the only sample rate multimon supports) and the rather peculiar incantation multimon needs to use standard input. Frequency scanning also works in conjunction with multimon.

## Network Streaming

Maybe you have a poor quality wifi link and don't want to send a full 1MS/s data stream around. Here is how to make a quick and dirty mp3 stream that will average 4KB-8KB/sec for voice channels. Other codecs may be swapped in as desired.

On the dongle side:

rtl_fm ..... | sox -traw -r24k -es -b16 -c1 -V1 - -tmp3 - | socat -u - TCP-LISTEN:8080

On the loudspeaker side:

netcat don.gle.ip.adr 8080 | play -t mp3 -


Q: Rtl_fm says it tunes to some other frequency instead of mine.

This is normal behavior, to dodge the DC spike present in E4000 tuners. The software tunes slightly off center and then corrects for the offset during demodulation.

Q: How to do skip channels when providing a range?

You don't. Use multiple ranges instead.

Q: How do I filter noise?

That is really hard. But Sox provides filters for it and you can use these filters in your play command.

Complete Reference

If you want to stream directly to a file, provide the file name as the final argument.

### -M fm | wbfm | am | usb | lsb | raw

### -N | -W | -M | -L | -U | -R

Selects a demodulator.

  • -N -M fm narrow band FM
  • -W -M wbfm wide band FM
  • -M -M am amplitude modulation
  • -L -M lsb lower sideband
  • -U -M usb upper sideband
  • -R -M raw raw output, 2x16 bit IQ pairs. Purely debugging and no sane person has any use for this.

-f <freq> | <start:stop:step>

Tune to a frequency. Value can be specified as an integer (89100000), a float (89.1e6) or as a metric suffix (89.1M). Multiple values or ranges will be scanned.

Scanning requires a non-zero squelch level.

-s <sample_rate>

Bandwidth of signal.

-d <index>

When using multiple dongles, this indicated which. You can also identify dongles by the text in the serial number field of the EEPROM

-g <gain>

A floating point gain value. Find legal values with rtl_test. The dongle will use the closest gain setting available.

-p <error>

Correct for the parts-per-million error in the crystal.

### -F 0 | 9

### -F

Enable a slightly higher quality downsampling FIR than the default boxcar filter. -F 0 is okay to use while -F 9 is still a work in progress.

### -A std | fast | lut

Choose how arctan is computed. Options select between the standard (floating) lib, a fast polynomial integer approximation and a precomputed look-up-table.

### -E dc | deemp | direct | offset

* -E dc dc blocking filter * -E deemp de-emphasis filter for WBFM * -E direct direct sampling mode * -E offset offset tuning mode (E4000 only)

Use multiple -E option to enable multiple features.

### -A

Enable a polynomial approximation of arctan. Fast and integer based.

### -D

Enables deemphasis filter used in WBFM.

Confusing Options

These kind of stink and are complicated and hard to explain.

### -E

Enables edge tuning. In the NBFM world, it is very common to specify the lower edge of the used bandwidth. For example, a NBFM transmitter operating with 20kHz bandwidth on 150MHz can be tuned with either

rtl_fm -E -f 150M -s 20k ... rtl_fm -f 150.010M -s 20k ...

By default this is disabled.

-l <squelch level>

Squelch is required for scanning multiple frequencies, to indicated when a frequency should be skipped. When used with a single frequency the output will be muted instead. Default value is 0/off.

Squelch values are extremely sensitive to external conditions (noise and antenna) as well as the other options used. For example, squelch should be halved when the sample rate is doubled. When in doubt, start at zero and work upwards.

### -o <oversampling>

With FM demodulation it is common to sample much faster than the actual bandwidth and downsample later to improve sound quality. This sets an integer multiple to scale the bandwidth. Default is 1 (no oversampling) but -o 4 is recommended. It has no effect on the output rate.

-t <squelch delay>

The number of squelch hits to count up to. Default value is 20. So the squelch must be trigger 20 times consecutively before rtl_fm moves on to the next channel.

Negative delay values cause rtl_fm to exit instead of hopping to another channel or muting the output. This is useful for scripts or trunked systems.

-r <output rate>

In some odd circumstances you might want to apply an extra low pass downsampling downsampling/resampling stage. This is for that, but most of the time it is rarely needed. It will also change the output rate. It is also a very poor quality resampler. Use sox for resampling