Files
urh/tests/data/FSK15.complex
Johannes Pohl e633a788ad Enhance accuracy of automatic interpretation (#550)
* rename signalFunctions -> signal_functions

* basic methods for auto interpretation

* cythonize k means

* reorder tests

* remove check and update gitignore

* estimate tolerance and implement score for choosing modulation type

* use absdiff

* remove comment

* cythonize messsage segmentation

* improve message segementation and add test for xavax

* integrate OOK special case

* add check if psk is possible

* integrate xavax and improve score

* improve noise detection

* add test for noise detection multiple messages

* improve noise detection

* homematic fix: use percetange of signal length instead of num flanks

* homematic has some trash at start of messages, which counts as flanks
* additonally set score to 0 if found only one bit length lower 5

* calculate minimum bit length from tolerance

* improve noise noise detection

* refactor limit and propose new limit calculation

* improve minimum bit length penalty

* only increase score for mod_type if bit length surpasses a minimum
* this way scoring loop later becomes easier and
* score is more accurate as there is no division needed which does
  not scale well with the length of message vectors

* remove demodulated complex files and demod live in tests

* remove enocean.coco

* add a new check to prevent PSK misclassification

* add tolerance unit test
* use z=2 for finding outlier free max in tolerance estimation

* prevent numpy warnings

* adapt threshold of unit test

* normalize the score by dividing by plateau vector length

* improve OOK segmentation:

Use minimum pulse length instead pause length for reference

* use 50 percentile for detecting n_digits in plateau rounding

* add elektromaten integration test

* improve center detection to deal with varying signal power levels
* use 10% clustering for rect signal
* calculate min and max of each cluster
* return max(minima) + min(maxima) / 2

* improve the center aggregation, separate between modulation types

* add validity checks if message can be ASK or FSK modulated

* use a weighted mean for center estimation: 60/40 for high/low

* improve bit length estimation: use decimal deviation for filtering

* add scislo test

* improve tolerance estimation: use 50 percentile + revert to normal mean for bitlength estimation

* add haar wavelet transform

* add median filter

* rename to Wavelet

* add signal generation with configurable snr for lab test

* add method for testdata generation

* prepare fsk test: generate messages and estimate parameters

* improve performance of plateau length filtering

* remove unused import

* improve robustness

* improve robustness

* add fsk error plot

* only append bit length if it surpasses minimum

* fix plot title

* improve noise level detection, prevent it from being too low

* integrate all modulations to test

* increase pause threshold for ook

* improve tolerance estimation

* improve noise detection: take maximum of all maxima of noise clusters

* improve scoring algorithm to prevent PSK misclassify as FSK

* use histogram based approach for center detection

* modulation detection with wavelets

* fix median filter when at end of data

* integrate modulation detection with wavelets

* improve robustness

* improve psk parameters

* improve psk threshold

* improve robustness

* swap psk angles for easier demod

* better xticks

* add message segmentation test and fix noise generation snr

* add error print

* update audi test

* fix runtime warning

* improve accuracy of center detection

* avoid warning

* remove unused functions

* fine tune fsk fft threshold

* update esaver test

* improve fsk fft threshold

* change test order

* update enocean test signal

* update enocean test signal

* enhance bit length estimation: use a threshold divisor histogram

* improve noise estimation: round to fourth digit

* update enocean signal

* consider special case if message pause is 0

* remove unused

* improve noise detection

* improve center detection

* improve center detection

* prevent warning

* refactor

* cythonize get_plateau_lengths

* improve syntax

* use c++ sort

* optimize PSK threshold

* optimize coverage

* fix buffer types

* integrate new auto detection routine

* update test

* remove unused stuff

* fix tests

* backward compat

* backward compat

* update test

* add threshold for large signals for performance

* update changelog

* make algorithm more robust against short bit length outliers

* make multi button for selecting auto detect options

* update unittest
2018-10-18 18:59:04 +02:00

62 KiB