Do It Yourself: Software-Defined Radio Digital DAB/DAB+ receiver

Transcription

1 Do It Yourself: Software-Defined Radio Digital DAB/DAB+ receiver Faculty of Computer Science, Electronics and Telecommunications Department of Telecommunications Delphi Technical Center, 30 VI

2 Team AGH University of Science and Technology Department of Telecommunications course/project: Software Defined Radio Practical Approach Tomasz P. Zieliński product manager, Matlab modelling, algorithms Jarosław Bułat program manager, software architect Michał Babiuch Viterbi, Reed-Solomon Ernest Biela time deinterleaver Szymon Dąbrowski FIC/MSC decoder workflow, ADTS superframe Piotr Jaglarz synchronizer workflow, NULL detection Adrian Karbowiak data decoder workflow, energy dispersal Sebastian Leśniak scheduler workflow Rafał Palej depuncturer, CRC Kacper Patro file handler, resampling Michał Rybczyński phase reference detection Dawid Rymarczyk data extraction from FIC Michał Rzepka FFT, demodulator workflow, dqpsk, freq. deinterleaver Krzysztof Szczęsny CLI client, WWW Wojciech Szmyd DC removal, AGC, audio decompressor and player Paweł Szulc data feeder workflow, resampling, Fc/Fs correction Jan Twardowski data extraction from packed mode Łukasz Wysogląd complexity analysis, testing Kacper Zych audio decoder workflow, manage output file

3 Agenda 1. Purpose of the project 2. Implementation: a) used tools b) program structure: general structure Scheduler DataFeeder Synchronizer & Demodulator DataDecoder AudioDecoder 3. Performance 4. Demonstration 5. Tasks for future polishing

4 Purpose of the project educational: to introduce issues of teamwork techniques and tools implementation: make reference C/C++ implementation of DAB+ standard utility: provide high-grade library (libsdrdab) under LGPL license, ready for integration with popular media players (vlc, mplayer, etc )

5 Workflow Knowledge base: DAB DAB Michael Höin, SW-Realisation of a DAB-Receiver with GNU Radio - MATLAB simulation prof. Tomasz P. Zieliński - extension of Höin s MATLAB code Development tools: GIT repository Eclipse CDT IDE under Linux googletest Framework Valgrind Distribution of work: 5 teams

6

7 Scheduler data flow Scheduler's main task is to keep whole system run and work properly. It manages method calls, creates buffers for data exchange between classes and manages its logic Set proper parameters for Demodulator, DataDecoder and AudioDecoder to correctly process data Keeps synchronization (null symbol position, carrier frequency drift) Calculates sampling frequency drift and exchange data between user and system There are 5 main classes in our system: DataFeeder writes data from dongle/file to buffer Synchronizer detects transmission mode and null symbol position, calculates carrier frequency drift Demodulator demodulates FIC and MSC from transmission frame DataDecoder decodes information from FIC and music stream from MSC AudioDecoder plays music, optionally saves it to file

8 Scheduler state machine To make system run continously Scheduler manages whole logic in state machine pattern. State diagram shows 5 main states: INIT - create and configure Synchronizer and DataFeeder SYNC - detect mode, prepare for decode FIC CONF - create Demodulator and DataDecoder, demodulate and decode FIC CONFSTATION - prepare for demodulate and decode specified station PLAY - decode FIC and MSC, play music

9 DataFeeder rtlsdr library challenges: changing parameters (frequency, gain) needs synchronous transfer data aquisition runs asynchronously the library isn t thread-safe customized branch Resampling: need to match DAB frame length libsamplerate (Secret Rabbit Code) available converters: SRC_SINC_BEST_QUALITY SRC_SINC_MEDIUM_QUALITY SRC_SINC_FASTEST SRC_LINEAR SRC_ZERO_ORDER_HOLD FC drift [Hz] - remodulate data (e j ) FS drift [ppm] - librtlsdr rtlsdr_set_freq_correction() function

10 Synchronizer DetectMode based on the distance between two successive NULL symbols Scheduler fills modeparameters structure according to the detected mode DetectAndDecodeNULL energy = sum of NULL size samples (window size) search for minimum energy in twice NULL size samples bad NULL quality -> look for minimum energy in frame size

11 Demodulator Map input buffer to OFDM symbols FFT samples per symbol + guard interval size 1 phase reference, 3 FIC, 72 MSC symbols for mode I fftw3 library in-place transform with pre-computed plan selected symbols only (chosen audio service) DQPSK phase shift calculation of selected OFDM symbols frequency deinterleaver deinterleaver table reshaping data to linear output buffer no temporary buffers

12 DataDecoder TimeDeinterleaver TimeDeinterleaver function copies audio bits to: provide redundancy allow data repair from defective TF It works: in stream mode with single copying procedure without extra buffers using vectors of indices to copy bits in data buffer It requires: access to multiple TF at the same time to shift the buffer every approaching TF Possibilities to do in the future: join with Depuncturer function

13 DataDecoder Data from FIG FIC carries basic information about: stations its organization and whole configuration There are not only information about subchannel organization, but also: programme type, which tells about context of particular service current date and time messages about changes in configuration labels describing station other services or messages like traffic congestions or emergency messages

14 DataDecoder Data from PADs FPAD - contains information about XPAD Problem with MPEG-2 and AAC FIG type 0 extension 1 In MPEG-2, FPAD is at the end of the frame, and XPAD is just before it and CRC In the AAC case FPAD is at the start of the frame CI in XPAD, Dynamic Label Multimedia Object Transfer

15 AudioDecoder Sinks for: speakers ogg files raw audio files unit tests GStreamer library: internal buffer occupancy upper/lower occupancy thresholds resampling

16 Performance Algorithm No interpolation With interpolation Viterbi: 64.0% 39% resampling: % remodulate: % Sch::Play: 1.61% 4.5% Sch::Process: 1.07% 4.15% ReadAsync: 7.83% 3.9% Audio: 11% 3.37% FFT: 1.71% 0.9% TimeDeinter: 0.32% 0.39% DetectMode: % DePuncturer: 0.43% 0.26% DeQpsk: 0.43% 0.26% NULL: 0.11% 0.13% perf record -a -e cycles --call-graph dwarf -F /Release/sdrdab-cli --katowice perf report --group -f

17 Demonstration 1. File-to-speaker 2. File-to-file 3. Air-to-speaker

18 Tasks for future polishing Implement missing algorithms: Firecode correction, PacketMode, etc.. Replace external libraries: resampling, ReedSolomon DAB+ encoder: real-time, USRP, distortion simulation (AWGN, NBI, fading/rayleigh channels, etc ) Support USRP and possible other hardware CPU/RAM optimalization Other platforms implementation: ARM (eg. RaspberryPI v2), Windows/MacOS GUI (standalone) + integration with VLC/mplayer/Kino More UnitTests: all modes, various signal conditions, various stream configurations Algorithms improvement: implement more convolutional decoders (hard-decision, BCJR, etc ) test and support Rayleigh channel (mobile case) decode many audio streams simultaneously New algorithms and tools: TEQ, FEQ, SNR calculation smart convolutional decoder (SNR dependent)