drm open source

8/9/2019 drm open source

1/22

Open-Source Implementation of a DigitalRadio Mondiale (DRM) Receiver

Dipl.-Ing. Alexander KurpiersDipl.-Ing. Volker Fischer

Darmstadt University of TechnologyInstitute of Communication Technology

Drea m


2/22

09/07/2003Open Source Implementation of a Digital Radio Mondiale

(DRM) Receiver2

Survey

Introduction & Motivation Channel Estimation

2D Interpolation 1D Wiener Interpolation in Time- and Frequency Direction

Synchronization Acquisition

Frequency: FFT-based Algorithm Time and Robustness Mode Detection: Guard-Interval Correlation DRM Frame Synchronization

Tracking Frequency and Sample Rate: Using Phase Increment of Pilots Time: Using Channel Estimation

Performance Simulation Conclusion and Outlook


3/22


(DRM) Receiver3

Motivation

DRM has a small bandwidth of less than 20 kHz - easyto handle with current PC sound cards

Real-time software implementation possible

No publicly available open-source DRM receiver

We looked for a test bed for OFDM algorithm development

(channel estimation, ICI compensation, synchronisation)Aim: complete DRM Receiver under GPLProject started in summer 2001

Digital Radio Mondiale (DRM) is a new OFDM-based digitalradio standard for the long-, medium- and short-wave ranges


4/22


(DRM) Receiver4

Introduction

OFDM

Vulnerable to frequency offsets (causes ICI) Timing critical if delay spread is in the range of the guard-

interval

Only one coefficient per carrier and symbol has to beestimated for equalisation


5/22

09/07/2003 Open Source Implementation of a Digital Radio Mondiale(DRM) Receiver

5

Software Overview

sound cardinterface

frequencyacquisition

sample ratecorrection

rob. modedetection

timingacquisitionuseful partextraction

frequencycorrection

OFDM

demodulationresample, freq.offset tracking

framesync.

channelestimation

timingtracking

sourcedecoders,

channeldecoders,OFDMdemux


6/22


6

Channel Estimation

Time

OF D M C

ar r i er I n d

ex

On everyOFDM carrier

the channeltransferfunctionhas to beestimated

lk H ,


7/22


7

Channel Estimation

Time

OF D M C

ar r i er I n d

ex

Scatteredpilots allow

for easychannelestimation


8/22


8

Channel Estimation

Time

OF D M C

ar r i er I n d

ex

twodimensional

WienerInterpolator isoptimal

but too

complex


9/22


9

Channel Estimation

Time

OF D M C

ar r i er I n d

ex

WienerInterpolator can

be separated intotwo 1D filters


10/22


10

Channel Estimation: Wiener Interpolation

MMSE solution: pRRh 1= p p ph

ph R : Cross-covariance matrix between and the noisy pilot estimatesh p

p pR : Auto-covariance matrix of the pilot estimates

IRRSNR

1

+= pp p p

Doppler profile of a typical shortwave channel:

Resulting correlation function:

2d

2

2

2d2

1)(

f

e f H

=

( ) ( )2

d2 k NT

f ek r d

=

Assuming uniform delay power spectrum with the length of theguard-interval:

( )

=

N N

llr Gsinc : Length of guard-intervalG N

N : Length of useful part


11/22


11

Channel Estimation in Frequency Direction

The following parameters were used in this simulation:Robustness mode B, 10 kHz bandwidth, 20 dB SNR, channel No. 3 (USConsortium)The mean squared error (MSE) between the estimated channel and an idealchannel estimation is plotted.

Blue line : Linearinterpolation

Green line :Windowed DFTalgorithm

Red line : Wienerinterpolation (using allpilot carriers for eachinterpolated cell

0 10 20 30 40-26

-24

-22

-20

-18

-16

-14

-12

Carrier Index

M S E [ d B ]


12/22


12

Frequency Acquisition (I)

Exploiting the power difference of the three frequency pilotcells and data cells Pilot cells: boosted, continuous tones Data cells: power spread due to modulation

FFT- based algorithm Squared norm of FFT calculated over more than one symbol (Estimation ofPSD)

Correlation with known frequency pilot positions

Effects of a large FFT window: Statistical properties of data cells more distinct, peak detection improved BUT fading channel effects reduce performance

3000 Hz

2250 Hz

750 Hz

k

l


13/22


13

Frequency Acquisition (II)

Placement of FFT window arbitraryNo prior timing information needed

Average error rate < 10% for all channels androbustness modes

Estimation of PSD

==

+

2

0),(

ac

sacq acmax

i

li pmm f R N f

f

21

0

2

,

ac

ac

=

+=

N

n

nm N

j

lnlm er R

Correlation with pilot positions


14/22


14

Time Acquisition (I)

Guard-interval correlation

Using energy in guard-interval- For multipath fading channel

( )+

=

=1

G )(G N

m

me

Example (idealised):

T N G0 t

Channel

is the resultingestimated timing position

{ })(maxarg G e

N G N Guard-interval

Usefulpart

t

G N 0 i

( )i

0

)(G e

G N

( ) ( )[ ]+

=

++ +=

122

*

G N i

in N nn N nn r r r r i


15/22


15

Time Acquisition (II)

Influence of Guard Energy Consideration on a two path fading channel:

Only correlationCorrelation with guard

energy consideration

Robustness modes can be detected by using time difference betweenpeaks (period equals useful part duration ) NT


16/22


16

Frame Synchronization

=

+

+=

1

0

2

1)(,2

2

)(,

1)(,)(,)(

T

t

G

t

t

t

L

ii pk

N N

j

i pk

i pk i pk zec

c zk

Time pilot pairs

Time

OF D M C

ar r i er I n

d ex

Assumption: channel is identical at adjacentpilot positions:

1)(,)(, +

i pk i pk t t H H

With this channel estimate we cancalculate the squared distance betweenreceived and pilot cells:

This yields a minimum at the beginningof the frame


17/22


17

Frequency Tracking

Frequency offset estimation based on phaseincrement between two successive symbols at the

frequency pilot carriers Frequency offset causes phase shift

( )( ) ( )( )= = +2

0 jacq*,acq,1s arg f f f z f zT j pl j pl

k l z , : Output of the FFT unit for the l -th symbol andthe k -th sub-carrier

sT : Duration of one symbol

)(f j p : Positions of frequency pilots


18/22

09/07/2003

Open Source Implementation of a Digital Radio Mondiale(DRM) Receiver

18

Time Tracking

{ }

=

=

1

0

2

,TiTr

TiTr

IFFT1)( N

ilik m H N k S

Using averaged IFFT-transformation of windowed channelestimation for estimation of channel impulse response

lk H ,

Afterwards using peak detection for first path estimation

{ })()(and,)(|min21 1 k S k S k S me mmm +> >=

=

10

min10

max

21

10,10max S S

- Definition of the bound

Estimated Impulse Response

t

k S

maxS

minS

1

2


19/22

09/07/2003


19

Software Modules

DataSource

DRMTransmitter

DRMChannel

DRMReceiver

UserInterface

SourceDecoder

Bit ErrorCounter

DownConverter

Antenna

Software can be used: together with Down Converter and Source Decoder to receive real-time DRM

radio broadcast- Source coding currently limited to plain MPEG4 AAC (no SBR, no CELP/HVXC)

for BER or Channel Estimation Simulations with build in Data Source,Transmitter and Channel Simulator


20/22


21/22

09/07/2003


21

Outlook

Developed DRM Receiver operates close to the possible limits Try to close the gap between ideal channel estimation and

realisation ICI compensation Decision directed channel estimation Noise cancellation for narrow-band interference

Software runs real-time on a 700 MHz Pentium PC

Improve to allow background reception Use SIMD instructions to speed up (MMX, SSE etc.) Improve pipelining of the algorithms to make acquisition phase

shorter

Source Decoder (faad2) needs additional features (SBR, CELP,HVXC)

See http://drm.sourceforge.net for details and download


22/22

09/07/2003


22

Downconverter

1. IF: 10.7 MHz

2. IF: 455 kHz3. IF: 5.25 kHz

Line-in48 kHz

DRM Program

Line-out48 kHz

USB

PC

Hard- and Software

Documents

drm open source