U NIVERSITY OF TURKU Modeling of DVB-H Link Layer Heidi Joki Deparment of Information Technology University of Turku Supervisor: Professor Jorma Virtamo Instructor: Jarkko Paavola, M.Sc.

U NIVERSITY OF TURKU Heidi Joki2 Agenda Background: Why was DVB-H developed? Services From DVB-T to DVB-H The DVB-H system DVB-H standards family Presentation of the DVB-H Link Layer Simulation model Simulation results New decoding algorithms Conclusions Further work

U NIVERSITY OF TURKU Heidi Joki3 Background: Why was DVB-H developed? There was a wish to bring TV-like services to mobile phones UMTS does not fulfil requirements for high bandwidth Internet applications, such as streaming video Mobile broadcasting is the best way to reach many users with reasonable cost DVB-T is not suitable for handheld battery powered devices

U NIVERSITY OF TURKU Heidi Joki4 Services Real time applications –TV broadcasting, info linked to events, games or quizzes Data carousel applications –Like teletext; stocks, weather, sports File Download –Buy newspaper, tourist map of city DVB-H in mobile phones => cellular network as return channel for interactivity, billing and authentication

U NIVERSITY OF TURKU Heidi Joki5 From DVB-T to DVB-H DVB-H is amendment of DVB-T for handheld devices Lower power consumtion in the receiver More flexibilyty in network planning Technical changes: –Time-slicing (Link layer) –MPE-FEC (Link layer) –4K OFDM mode (Physical layer) –IP datacast (Network layer) –Signaling

U NIVERSITY OF TURKU Heidi Joki6 The DVB-H system

U NIVERSITY OF TURKU Heidi Joki7

U NIVERSITY OF TURKU Heidi Joki8 Presentation of the DVB-H Link Layer Link Layer Packets (TX) Time-Slicing MPE-FEC Reed-Solomon(255,191) MPE- and FEC-sections Transport Stream Section parsing and Decapsulation (RX) Erasure Decoding (RX)

U NIVERSITY OF TURKU Heidi Joki9 Link Layer Packets (transmitter)

U NIVERSITY OF TURKU Heidi Joki10 Data sent in bursts, one burst per MPE-FEC frame Enables power saving (≤90%) Delta-t, time to start of next burst, is announced in the section header No separate synchronization needed; Receiver clock has to be stable only until next burst Supports use of receiver for network monitoring during off-periods Time-slicing

U NIVERSITY OF TURKU Heidi Joki11 MPE-FEC in TX (1/2) IP header (20B)Payload (0-1480B) Last punctured RS column. First punctured RS column Parity bytes in last FEC section. Parity bytes carried in section 2 Parity bytes carried in section 1 Last data padding column. First data padding column Last IP datagram Padding bytes. 2nd IP dg cont.. 3rd IP dg 1st IP dg cont. 2nd IP datagram 1st IP datagram 1 1 Nbr of rows 256, 512, 786 or Application data tableRS data table

U NIVERSITY OF TURKU Heidi Joki12 MPE-FEC in TX (2/2) Max 1500B IP datagrams (as Ethernet) IP datagrams encapsulated column-wise into the Application Data Table (ADT) ADT encoded row-wise with RS(255,191) Virtual interleaving is achieved! Code shortening and puncturing used for achieving different MPE-FEC code rates Different number of rows in MPE-FEC frame give different burst sizes Number of rows and the use of MPE-FEC is signalled to the receiver

U NIVERSITY OF TURKU Heidi Joki13 Reed-Solomon(255,191) Hamming distance d = n-k+1 = 65 Correction capabillity –t u = 32 errors if pure error correction used –t e = 64 erasures if pure erasure correction used Hamming distance depends on the amount of transmitted RS columns

U NIVERSITY OF TURKU Heidi Joki14 MPE- and MPE-FEC sections IP datagrams form payload of MPE- sections RS data columns form payload of MPE- FEC sections 12B section header added CRC-32 calculated and 4 redundancy bytes placed at the end of the section CRC-32 is used for error detection in the receiver

U NIVERSITY OF TURKU Heidi Joki15 MPE section headerMPE-FEC section header SyntaxbitsSyntaxbits table_id8 8 section_syntax_indicator1 1 private_indicator1 1 reserved2Reserved2 section_length12section_length12 MAC_address_68padding_columns8 MAC_address­_58reserved_for_future_use8 reserved2Reserved2 payload_scrambling_control2reserved_for_future_use5 address_ scrambling_control2 LLC_snap_flag1 current_next_indicator1 1 section_number8 8 last_section_number8 8 Real_time_parameters32Real_time_parameters32

U NIVERSITY OF TURKU Heidi Joki16 Real time parameters Delta-t = time to beginning of next burst Table_bounary = ’1’ for last section of ADT or RS data table Frame_boundary = ’1’ for last section of a MPE-FEC frame Address = number of cell in the MPE-FEC frame for the first byte of the payload carried in that section

U NIVERSITY OF TURKU Heidi Joki17 Transport Stream TS packet = 4B TS header + 184B payload 13 bit PID in the TS header indicates Elementary Stream and data type transport_error_indicator (1 bit) set to ’1’ by physical layer RS(204,188) decoder in the receiver if error correction failed MPE-FEC header (12B)Column (max 1024B)CRC-32 (4B) MAC sublayer: MPE and MPE-FEC sections (Header includes 4B Real time parameters) MPE header (12B)IP datagramCRC-32 (4B) TS Header (4B)Payload (184B)TS Header (4B)Payload (184B)... MPEG-2 Transport Stream

U NIVERSITY OF TURKU Heidi Joki18 Section parsing and decapsulation in the Receiver RX receives TS with a certain PID Find first byte of the section –table_id = 62 (MPE) or 120 (FEC) Find section length Do CRC-32 check –OK -> find address and decapsulate the section payload into the frame –Failed -> mark bytes as erasures

U NIVERSITY OF TURKU Heidi Joki19 Erasure decoding in DVB-H Erasure Info Table (EIT) of same size as MPE-FEC frame ’0’ = reliable byte, ’1’ = erasure If a section fails CRC-32 check, the complete datagram/RS column is marked as ’erasure’ RS decoder can correct 64 erasures/row if all RS columns are transmitted

U NIVERSITY OF TURKU Heidi Joki20 Simulation model of Finnish WingTV consortium

U NIVERSITY OF TURKU Heidi Joki21 Simulation model: motivation The number of link layer and physical layer parameters add up to 14400! Simulation is the fastest and most economic way of evaluating the impact of different parameters Simulation provides an opportunity to test new ways of parsing, decapsulation and decoding ParameterOptionsExplanation Modulation3QPSK, 16QAM, 64QAM FFT-size32K, 4K, 8K In-depth interleaver2On / Off (only for 2K and 4K) Guard Interval41/4, 1/8, 1/16, 1/32 CC rate51/2, 2/3, 3/4, 5/6, 7/8 MPE-FEC code rate61/2, 2/3, 3/4, 5/6, 7/8, 1 Burst size4256, 512, 768, 1024 rows Burst bit rate2 Number of combinations 14400

U NIVERSITY OF TURKU Heidi Joki22 Simulation model (link layer) Outside the scope of the DVB-H standard, means for TS erasure decoding and hierarchical decapsulation were also implemented (not included in the figure).

U NIVERSITY OF TURKU Heidi Joki23 TS erasure decoding Except the CRC erasure decoding, means for TS erasure decoding was implemented Symbols in the MPE-FEC frame are marked as reliable or unreliable based on the transport_error_indicator in the TS header IP datagram lengths not considered

U NIVERSITY OF TURKU Heidi Joki24 The error pattern MPEG-2 Source DVB-T Modulator Channel Simulator Noise Generator DVB-T/H Receiver MPEG-2 Test Signal Different DVB- T modes Hardware channel simulator and noise generator: COST 207 TU channel Fd C/N Logic Analyzer Only the TS error statistics were saved into the file TS error Data: … Provided by Nokia

U NIVERSITY OF TURKU Heidi Joki25 Simulation parameters The effect of the following parameters on the MPE-FEC FER can be examined: Burst size, i.e. number of rows in MPE-FEC frame MPE-FEC code rate Length of IP datagrams FEC decoder type: TS erasure decoding vs. CRC erasure decoding The length of the burst, i.e. the interleaving length The above mentioned parameters can be simulated with the following physical channel parameters: Modulation Doppler frequency Convolutional code rate Channel model: TU6, indoor, pedestrian, etc.

U NIVERSITY OF TURKU Heidi Joki26 Performed simulations The simulations were performed with 256- and 1024-row frames IP datagram length was 1500 bytes Two different simulations were carried out –CRC erasure decoding –TS erasure decoding The aim was to compare the two different methods and to study the amount of unnecessary erasures added to the EIT by the CRC decoding Channel model:TU6 Modulation:16 QAM Doppler frequency:10 Hz CC rate:½ Amount of TS packets: Amount of TS data:788 MB IP datagram length:1500 Bytes Amount of IP data:256 rows: 560 MB 1024 rows: 570 MB MPE-FEC code rate:¾ Signal to noise ratio:17 – 20 dB Amount of MPE-FEC frames: 256 rows: frames 1024 rows: 2927 frames

U NIVERSITY OF TURKU Heidi Joki27 CRC erasure decoding vs. TS erasure decoding EIT64The RS decoder, using erasure information, is able to correct 64 bytes of CRC-32 erasure data per row in an MPE-FEC frame. Real 32The RS decoder is able to correct 32 erroneous bytes per row. The error locations are unknown. Errors are lost TS packets. The length of the IP datagram is ignored. Real 64The RS decoder, using erasure information, is able to correct 64 erroneous bytes per row. Errors are lost TS packets. The length of the IP datagram is ignored.

U NIVERSITY OF TURKU Heidi Joki28 Symbol error ratio using CRC erasure decoding Input SER equals TS PER. All symbols in an erroneous TS packet are considered incorrect. Output SER is the SER after CRC erasure decoding using RS(255,191)

U NIVERSITY OF TURKU Heidi Joki29 Result analysis CRC-32 erasure decoding adds far too many unnecessary erasures. When transmitting 1500B IP datagrams in the smallest frame, the gain of using FEC is almost lost if using erasures based on CRC-32 TS erasure decoding saves gain in all simulations Using a larger MPE-FEC frame gives improvement in gain, when burst length is not considered.

U NIVERSITY OF TURKU Heidi Joki30 Drawbacks of the DVB-H standard CRC adds too much erasures into EIT Lack of protection of the section header Standard length of IP datagrams or MPE sections preferable than various length –Achieving constant TS bit rate (or almost constant for streaming video) –Decapsulation possible, though section header is lost Not 100% certainity of ’reliable’ bytes in MPE- FEC frame has to be considered

U NIVERSITY OF TURKU Heidi Joki31 Suggestions for improvements (without changing the standard) TX: Introducing standard length of IP datagrams (e.g. 1 or 2 columns) RX: Using TS erasure decoding based on the transport_error_indicator in the TS header RX: Using hierarchical decapsulation and decoding if needed (also decapsulate erroneous packets, most of it is probably correct!) RX: Using combination of erasure and error decoding

U NIVERSITY OF TURKU Heidi Joki32 The algorithm for hierarchical decapsulation and hierarchical decoding 1.Perform hierarchical decapsulation of TS packets, using the transport_error_indicator when filling in the erasure info table (EIT). Lost data is market with ‘1’, decapsulated but unreliable data is marked with ‘2’ and correct data with ‘0’ in the EIT. 2.Consider all unreliable bytes, marked with ‘1’ or ‘2’ in the EIT, as erasures. 3.If the amount of unreliable bytes is less than 64, use the remaining Hamming distance for error decoding. Perform the erasure (and error) decoding. 4.If the amount of unreliable bytes exceeds 64, consider the bytes marked with ‘2’ in the EIT as reliable and repeat step 3. The pure erasure decoding could also fail if some of the bytes marked as reliable are erroneous. In this case step 4 is useful, since it might leave some more Hamming distance for error correction. This algorithm can be combined with CRC or TS erasure decoding. TS erasure decoding is recommended.

U NIVERSITY OF TURKU Heidi Joki33 Further work on the simulator Means for the user to input the simulation parameters should be implemented. At least the following parameters should be read: –MPE-FEC code rate –The names of the IP data and error pattern files –Burst size and duration –Decoding method to be used; TS erasure or CRC erasure correction The TS erasure decoding should be implemented so that IP datagram lengths are taken into account. Also combinations of erasure and error correction should be thought of Time-slicing should be implemented Besides the FER, the output of the simulator should include IP data along with erasure information, which is used by a potential RS decoder at the application layer The simulator should be able to handle a multiplex of many elementary streams Hierarchical decapsulation and decoding should be implemented A symbol based TS error pattern is needed Functions should be optimized for shortening the simulation time

U NIVERSITY OF TURKU Heidi Joki34 Future work on DVB-H link layer and physical layer The impact of the IP datagram lengths and the MPE-FEC code rates should be studied carefully The decoding process should be improved and different decoding algorithms should be studied Finding the best means of decapsulation and decoding using all received data is already quite a challenge. However, the receiver manufacturers would probably profit from implementing solutions for decoding based on a combination of TS erasure and error correction. Proper channel models for indoor and pedestrian use cases should be developed Based on the channel models, error patterns based on symbol or bit errors could be developed on TS level

U NIVERSITY OF TURKU Heidi Joki35 Thank You! Questions? For more information contact