airspy-fmradion

• Version 20210303-0
• For MacOS and Linux

Contributing

See CONTRIBUTING.md for the details.

No more semantic versioning

The semantic versioning scheme of airspy-fmradion has utterly failed.

New version number scheme: YYYYMMDD-N (N: subnumber, starting from 0, unsigned integer)

Known issues and changes

• MacOS build is tested with 10.15.7 Catalina with Xcode 12.4 Command Line Tools. MacOS Big Sur 11.x versions haven't been tested yet.
• For Raspberry Pi 3 and 4, Airspy R2 10Mbps and Airspy Mini 6Mbps sampling rates are not supported due to the hardware limitation. Use in 2.5Mbps for R2, 3Mbps for Mini.
• FM ppm display shows ppb (0.001ppm) digits since 20210206-0.
• FM Pilot PLL is under revision and reconstruction. Initial analysis result is available at doc/fm-pll-filtereval.py (requires Python 3, SciPy, matplotlib, and NumPy).
• The 2nd-order LPF of FM Pilot PLL is applied twice since 20210116-0.
• Timestamp file format has been changed since 20201204-0.
• PortAudio is required since Version 20201023-0. Use PortAudio v19. Former ALSA output driver is replaced by more versatile PortAudio driver, which is compatible both for Linux and macOS.
• libvolk is required since v0.8.0. If you don't want to install libvolk, use v0.7.8 instead. Use the latest master branch of libvolk. Configure the volk_config file with volk_profile -b for the maximum performance. See INSTALL-latest-libvolk.md for the details.
• v0.8.5 and the earlier versions set the compilation flag of -ffast-math, which disabled the processing of NaN. This will cause a latch-up bug when the multipath filter coefficients diverge. Removed -ffast-math for the stable operation.
• v0.9.0-test1 to v0.9.5 had calculation error due to volk_32f_expfast_32f() in IfAgc::process() method. Fixed this by replacing to the more accurate calculation code of volk_32f_exp_32f().

What is airspy-fmradion?

• airspy-fmradion is a software-defined radio receiver for FM and AM broadcast radio, and also NBFM/DSB/USB/LSB/CW/WSPR utility communications, specifically designed for Airspy R2, Airspy Mini, Airspy HF+, and RTL-SDR.

What does airspy-fmradion provide?

• Mono or stereo decoding of FM broadcasting stations
• Mono decoding of AM stations
• Decoding NBFM/DSB/USB/LSB/CW/WSPR stations
• Buffered real-time playback to soundcard or dumping to file
• Command-line interface (only)

Usage

airspy-fmradion -t airspy -q \
    -c freq=88100000,srate=10000000,lgain=2,mgain=0,vgain=10 \
    -b 1.0 -P -

airspy-fmradion -t airspyhf -q \
    -c freq=88100000,srate=768000 \
    -b 1.0 -R - | \
    play -t raw -esigned-integer -b16 -r 48000 -c 2 -q -

airspy-fmradion -m am -t airspyhf -q \
    -c freq=666000 \
    -b 0.5 -F - | \
    play --buffer=1024 -t raw -e floating-point -b32 -r 48000 -c 1 -q -

airspy-fmradion requires

• Linux / macOS
• C++11 (gcc, clang/llvm)
• Airspy library
• Airspy HF library
• RTL-SDR library
• sndfile
• The SoX Resampler library aka libsoxr
• VOLK
• PortAudio
• Tested: Airspy R2, Airspy Mini, Airspy HF+ Dual Port, RTL-SDR V3
• Fast computer
• Medium-strong radio signals

For the latest version, see https://github.com/jj1bdx/airspy-fmradion

Recommended utilities

• sox

Branches and tags

• Official releases are tagged
• master is the "production" branch with the most stable release (often ahead of the latest release though)
• dev is the development branch that contains current developments that will be eventually released in the master branch
• Other branches are experimental (and presumably abandoned)

Prerequisites

Airspy HF+ firmware

Use the latest version of Airspy HF+ firmware, available at Airspy HF+ Dual Port and Airspy HF+ Discovery Web pages.

airspy-fmradion sets the default sampling rates to 384kHz for FM broadcast, and 192kHz for the other modes. Old Airspy HF+ firmwares do not support the lower sampling rate other than 768kHz.

Required libraries

Note: the master branch of libvolk is now required from v0.8.1.

If you install from source in your own installation path, you have to specify the include path and library path. For example if you installed it in /opt/install/libairspy you have to add -DAIRSPY_INCLUDE_DIR=/opt/install/libairspy/include -DAIRSPYHF_INCLUDE_DIR=/opt/install/libairspyhf/include to the cmake options.

Debian/Ubuntu Linux

• sudo apt-get install cmake pkg-config libusb-1.0-0-dev libasound2-dev libairspy-dev libairspyhf-dev librtlsdr-dev libsoxr-dev libsndfile1-dev portaudio19-dev

macOS

• Install HomeBrew libraries as in the following shell script
• See https://github.com/pothosware/homebrew-pothos/wiki
• Use HEAD for airspy and airspyhf

brew tap pothosware/homebrew-pothos
brew tap dholm/homebrew-sdr #other sdr apps
brew update
brew install portaudio
brew install libsoxr
brew install libsndfile
brew install rtl-sdr
brew install airspy --HEAD
brew install airspyhf --HEAD

Install the latest libvolk

Install libvolk as described in INSTALL-latest-libvolk.md.

Dependency installation details

libvolk

• See GitHub gnuradio/volk repository for the details.
• libvolk requires Boost to compile.

libairspyhf

Use the latest HEAD version.

libairspy

Note: this is applicable for both macOS and Linux.

Install and use the latest libairspy --HEAD version for:

• Working airspy_open_devices(), required by airspy_open_sn(). See this commit for the details.
• Proper transfer block size. if_blocksize for Airspy HF+ is reduced from 16384 to 2048, following this commit.

Installing

To install airspy-fmradion, download and unpack the source code and go to the top level directory. Then do like this:

• mkdir build
• cd build
• cmake ..

CMake tries to find librtlsdr. If this fails, you need to specify the location of the library in one the following ways:

cmake .. \
  -DAIRSPY_INCLUDE_DIR=/path/airspy/include \
  -DAIRSPY_LIBRARY_PATH=/path/airspy/lib/libairspy.a
  -DAIRSPYHF_INCLUDE_DIR=/path/airspyhf/include \
  -DAIRSPYHF_LIBRARY_PATH=/path/airspyhf/lib/libairspyhf.a \
  -DRTLSDR_INCLUDE_DIR=/path/rtlsdr/include \
  -DRTLSDR_LIBRARY_PATH=/path/rtlsdr/lib/librtlsdr.a

PKG_CONFIG_PATH=/path/to/airspy/lib/pkgconfig cmake ..

Compile and install

• make -j4 (for machines with 4 CPUs)
• make install

Basic command options

• -m devtype is modulation type, one of fm, nbfm, am, dsb, usb, lsb, cw, wspr (default fm)
• -t devtype is mandatory and must be airspy for Airspy R2 / Airspy Mini, airspyhf for Airspy HF+, rtlsdr for RTL-SDR, and filesource for the File Source driver.
• -q Quiet mode.
• -c config Comma separated list of configuration options as key=value pairs or just key for switches. Depends on device type (see next paragraph).
• -d devidx Device index, 'list' to show device list (default 0)
• -M Disable stereo decoding
• -R filename Write audio data as raw S16_LE samples. Use filename - to write to stdout
• -F filename Write audio data as raw FLOAT_LE samples. Use filename - to write to stdout
• -W filename Write audio data to .WAV file
• -P device_num Play audio via PortAudio device index number. Use string - to specify the default PortAudio device
• -T filename Write pulse-per-second timestamps. Use filename '-' to write to stdout
• -b seconds Set audio buffer size in seconds (default: 1 second)
• -X Shift pilot phase (for Quadrature Multipath Monitor) (-X is ignored under mono mode (-M))
• -U Set deemphasis to 75 microseconds (default: 50)
• -f Set Filter type
  • for FM: wide and default: none, medium: +-156kHz, narrow: +-121kHz
  • for AM: wide: +-9kHz, default: +-6kHz, medium: +-4.5kHz, narrow: +-3kHz
  • for NBFM: wide: +-20kHz, default: +-10kHz, medium: +-8kHz, narrow: +-6.25kHz
• -l dB Enable IF squelch, set the level to minus given value of dB
• -E stages Enable multipath filter for FM (For stable reception only: turn off if reception becomes unstable)
• -r ppm Set IF offset in ppm (range: +-1000000ppm) (Note: this option affects output pitch and timing: use for the output timing compensation only!

Major changes

Timestamp file format

• For FM: pps_index sample_index unix_time if_level
• For the other modes: block unix_time if_level
• if_level is in dB

Rate compensation for adjusting audio device playback speed offset

• Background: some audio devices shows non-negligible offset of playback speed, which causes eventual audio output buffer overflow and significant delay in long-term playback.
• How to fix: compensating the playback speed offset gives more accurate playback timing, by sacrificing output audio pitch accuracy. A proper compensation will eliminate the cause of increasing output buffer length, by sending less data (lower sampling rate) to the conversion process.
• You can specify the compensation rate by ppm using -r option.
• How to estimate the rate offset: when elapsed playback time is Tp [seconds] and output buffer length (buf= in the debug output) increases during the time is Ts [seconds], the compensation rate is (Ts/Tp) * 1000000 [ppm].
• For example, if the output buffer length increases for 1 second after playing back for 7 hours (25200 seconds), the offset rate is 1/25200 * 1000000 ~= 39.68ppm.
• +- 100ppm offset is not uncommon among the consumer-grade audio devices.
• +- 100ppm pitch change may not be recongizable by human.

Caveats for the rate compensation

• Do not use this feature if the per-sample accuracy is essential.
• Do not use this feature for non-realtime output (for example, to files).
• Output audio pitch increases as the offset increases.
• Too much compensation will cause output underflow.
• This feature causes fractional (non-integer) resampling by IfResampler class, which causes more CPU usage.

Smaller latency

• v0.9.2 uses smaller latency algorithms for all modulation types and filters. The output frequency characteristics may be different from the previous versions.

Audio gain adjustment

• Since v0.4.2, output maximum level is back at -6dB (0.5) (adjust_gain() is reintroduced) again, as in pre-v0.2.7
• During v0.2.7 to v0.4.1, output level was at unity (adjust_gain() is removed)
• Before v0.2.7, output maximum level is at -6dB (0.5)

Audio and IF downsampling is now performed by libsoxr

• Output of the stereo decoder is downsampled by libsoxr to 48kHz
• Quality: SOXR_VHQ
• 19kHz cut LPF implemented for post-processing libsoxr output
• Do not use SOXR_STEEP_FILTER because it induces unacceptable higher latency
• Audio sample rate is fixed to 48000Hz

Phase discriminator uses GNU Radio fast_atan2f()

• From v0.7.8-pre0, GNU Radio fast_atan2f() which has ~20-bit accuracy, is used for PhaseDiscriminator class and the 19kHz pilot PLL.
• The past fastatan2() used in v0.6.10 and before was removed due to low accuracy (of ~10 bits)
• Changing from the past atan2() to fast_atan2f() showed no noticeable difference of the THD+N (0.218%) and THD (0.018%). (Measured from JOBK-FM NHK Osaka FM 88.1MHz hourly time tone 880Hz, using airwaves after the multipath canceler filter of -E36)
• The past fastatan2() allowed +-0.005 radian max error
• libm atan2() allows only approx. 0.5 ULP as the max error for macOS 10.14.5, measured by using the code from "Error analysis of system mathematical functions " by Gaston H. Gonnet (1 ULP for macOS 64bit double = 2^(-53) = approx. 10^(15.95))

FM multipath filter

• A Normalized LMS-based multipath filter can be enabled after IF AGC
• IF sample stages can be defined by -E options
• Reference amplitude level: 1.0
• For Mac mini 2018 with 3.2 GHz Intel Core i7, 288 stages consume 99% of one CPU core
• This filter is not effective when the IF bandwidth is narrow (192kHz)
• The multipath filter starts after discarding the first 100 blocks. This change is to avoid the initial instability of Airspy R2.
• Note: this filter recalculates the coefficients for every four (4) samples, to reduce the processing load.

Multipath filter configuration

• v0.7.3 and later: -E36 for 108 previous and 36 after stages (ratio 3:1). The multipath filter order: (4 * stages) + 1
• For reference only: v0.7.3-pre1 and before: -E72 for 72 previous and 72 after stages (ratio 1:1), summary: set the -E parameter to 1/2 of the previous value
• Rule of thumb: -E36 is sufficient for a stable strong singal (albeit with considerable multipath distortion).

FM L-R signal boosted for the stereo separation improvement

• Teruhiko Hayashi suggested boosting L-R signal by 1.017 for a better stereo separation. Implemented since v0.7.6-pre3.
• DiscriminatorEqualizer removed since v1.7.6-pre3 (needs more precise compensation, presumably with an FIR filter.

FM deemphasis error prevention

• Teruhiko Hayashi suggested applying deemphasis before the sampling rate conversion, at the demodulator rate, higher than the audio output rate. Implemented since v0.7.6.

No-goals

• CIC filters for the IF 1st stage (unable to explore parallelism, too complex to compensate)
• Using lock-free threads (boost::lockfree::spsc_queue didn't make things faster, and consumed x2 CPU power)

Filter design documentation

General characteristics

• Filter is implemented as the libsoxr sampling converter
• Filter cutoff by libsoxr: 0.982 * sampling frequency

For FM

• FM Filter coefficients are listed under doc/filter-design

For NBFM

• Deviation: normal +-8kHz, for wide +-17kHz
• Output audio LPF: flat up to 4kHz
• NBFM Filter coefficients are listed under doc/filter-design
• default filter width: +-10kHz
• Narrower filters by -f options: middle +-8kHz, narrow +-6.25kHz
• Wider filters by -f options: wide +-20kHz (with wider deviation of +-17kHz)
• Audio gain reduced by -3dB to prevent output clipping

For AM and DSB

• AM Filter coefficients are listed under doc/filter-design
• default filter width: +-6kHz
• Narrower filters by -f options: middle +-4.5kHz, narrow +-3kHz
• Wider filters by -f options: wide +-9kHz

For SSB (USB/LSB)

• For USB: shift down 1.5kHz -> LPF -> shift up 1.5kHz
• For USB: shift up 1.5kHz -> LPF -> shift down 1.5kHz
• Rate conversion of 48kHz to 12kHz and vice versa for the input and output of LPF
• 12kHz sampling rate LPF: flat till +-1200Hz, -3dB +-1320Hz, -10dB +-1370Hz, -58.64dB at +-1465Hz

For CW

• Zeroed-in pitch: 500Hz
• Filter width: +- 100Hz flat, +-250Hz for cutoff
• Uses downsampling to 12kHz for applying a steep filter

For WSPR

• Set USB TX carrier frequency for reception (as shown in WSPRnet)
• Pitch: 1500Hz
• Filter width: +- 100Hz flat, +-250Hz for cutoff
• Uses downsampling to 12kHz for applying a steep filter

AGC algorithm

• Use simple logarithm-based AGC algorithm, which only depends on the single previous sample
• See https://www.mathworks.com/help/comm/ref/comm.agc-system-object.html for the implementation details
• IF AGC: gain up to 100dB (100000) (for broadcast FM: 80dB (10000))
• Audio AGC: gain up to 7dB (5.0)

Airspy R2 / Mini modification from ngsoftfm-jj1bdx

The modification strategy

Twitter @lambdaprog suggested the following strategy:

Try starting with 10MSPS and that small conversion filter (7 taps vs. the standard 47 taps), then decimate down to approx. 312.5 ksps (decimation by 32), then feed the FM demod. The overall CPU usage will be very low and the bit growth will give 14.5 bit resolution.

Feature changes

• Halfband kernel filter designed by Twitter @lambdaprog is set for Airspy conversion filter
• Finetuner is removed (Not really needed for +-1ppm or less offset)

Airspy R2 / Mini configuration options

• freq=<int> Desired tune frequency in Hz. Valid range from 1M to 1.8G. (default 100M: 100000000)
• srate=<int> Device sample rate. list lists valid values and exits. (default 10000000). Valid values depend on the Airspy firmware. Airspy firmware and library must support dynamic sample rate query.
• lgain=<x> LNA gain in dB. Valid values are: 0, 1, 2, 3, 4, 5, 6, 7, 8 ,9 ,10, 11 12, 13, 14, list. list lists valid values and exits. (default 8)
• mgain=<x> Mixer gain in dB. Valid values are: 0, 1, 2, 3, 4, 5, 6, 7, 8 ,9 ,10, 11 12, 13, 14, 15, list. list lists valid values and exits. (default 8)
• vgain=<x> VGA gain in dB. Valid values are: 0, 1, 2, 3, 4, 5, 6, 7, 8 ,9 ,10, 11 12, 13, 14, 15, list. list lists valid values and exits. (default 0)
• antbias Turn on the antenna bias for remote LNA (default off)
• lagc Turn on the LNA AGC (default off)
• magc Turn on the mixer AGC (default off)

Airspy HF+ modification from airspy-fmradion v0.2.7

Sample rates and IF modes

• This version supports 768kHz zero-IF mode and 384/256/192kHz low-IF modes

Conversion process for 768kHz zero-IF mode

• LPFIQ is single-stage
• IF center frequency is down Fs/4 than the station frequency, i.e: when the station is 76.5MHz, the tuned frequency is 76.308MHz
• Airspy HF+ allows only 660kHz alias-free BW, so the maximum alias-free BW for IF is (660/2)kHz - 192kHz = 138kHz

Conversion process for 384/256/192kHz low-IF mode

• IF center frequency is not shifted

Airspy HF configuration options

• freq=<int> Desired tune frequency in Hz. Valid range from 0 to 31M, and from 60M to 240M. (default 100M: 100000000)
• srate=<int> Device sample rate. list lists valid values and exits. (default 384000). Valid values depend on the Airspy HF firmware. Airspy HF firmware and library must support dynamic sample rate query.
• hf_att=<int> HF attenuation level and AGC control.
  • 0: enable AGC, no attenuation
  • 1 - 8: disable AGC, apply attenuation of value * 6dB

RTL-SDR

Sample rate

• Valid sample rates are from 1000000 to 1250000 [Hz]
• The default value is 1200000Hz

RTL-SDR configuration options

• freq=<int> Desired tune frequency in Hz. Accepted range from 10M to 2.2G. (default 100M: 100000000)
• gain=<x> (default auto)
  • auto Selects gain automatically
  • list Lists available gains and exit
  • <float> gain in dB. Possible gains in dB are: 0.0, 0.9, 1.4, 2.7, 3.7, 7.7, 8.7, 12.5, 14.4, 15.7, 16.6, 19.7, 20.7, 22.9, 25.4, 28.0, 29.7, 32.8, 33.8 , 36.4, 37.2, 38.6, 40.2, 42.1, 43.4, 43.9, 44.5, 48.0, 49.6
• srate=<int> Device sample rate. valid values in the [225001, 300000], [900001, 3200000] ranges. (default 1000000)
• blklen=<int> Device block length in bytes (default RTL-SDR default i.e. 64k)
• agc Activates device AGC (default off)
• antbias Turn on the antenna bias for remote LNA (default off)

File Source driver

• Reads an IQ signal format file and decode the output.
• This device is still experimental.

File Source configuration options

• freq=<int> Frequency of radio station in Hz.
• srate=<int> IF sample rate in Hz.
• filename=<string> Source file name. Supported encodings: FLOAT, S24_LE, S16_LE
• zero_offset Set if the source file is in zero offset, which requires Fs/4 IF shifting.
• blklen=<int> Set block length in samples.
• raw Set if the file is raw binary.
• format=<string> Set the file format for the raw binary file. Supported formats: U8_LE, S8_LE, S16_LE, S24_LE, FLOAT

Authors and contributors

• Joris van Rantwijk, primary author of SoftFM
• Edouard Griffiths, F4EXB, primary author of NGSoftFM (no longer involving in maintaining NGSoftFM)
• Kenji Rikitake, JJ1BDX, maintainer
• András Retzler, HA7ILM, for the former AF/IF AGC code in csdr
• Youssef Touil, Airspy Founder, aka Twitter @lambdaprog, for the intriguing exchange of Airspy product design details and the technical support
• Iowa Hills Software, for their FIR and IIR filter design tools
• Brian Beezley, K6STI, for his comprehensive Web site of FM broadcasting reception expertise and the idea of Quadrature Multipath Monitor
• Ryuji Suzuki, for reviewing the FM multipath filter coefficients and suggesting putting more weight on picking up more previous samples from the reference point than the samples after
• Teruhiko Hayashi, JA2SVZ, the creator of FM FPGA Tuner popular in Japan, for reviewing the measurement results of FM broadcast reception of airspy-fmradion, and various constructive suggestions
• Takehiro Sekine, for suggesting using GNU Radio's VOLK for faster calculation, and implementing Filesource device driver

License

• As a whole package: GPLv3 (and later). See LICENSE.
• csdr AGC code: BSD license.
• Some source code files are stating GPL "v2 and later" license.

Repository history

• This repository is forked from ngsoftfm-jj1bdx 0.1.14 and merged with airspfhf-fmradion.