JP2007335926A - Method of generating parity for fec, and transmitter - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem wherein a large buffer memory is required when using an error correction code for generating an error correction code in large units as in interleaved Reed-Solomon, and packets are sent onto a network in groups and hence a probability of packet loss is high. <P>SOLUTION: There are provided: a parity generation means for generating K-bytes parity data for each N-bite unit of input data; a selection means for selecting one of the input data and the parity data for output; a division means for dividing the output of the selection means in M-byte units; and an RTP packet generation means for adding a header to the divided data for outputting an RTP packet. In this case, N and K are integral multiples of M, and the selection means has a communication means for sending N-bytes input data to the RTP packet generation means before sending the K-bytes parity data for outputting the RTP packet to the network at a prescribed interval. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a parity generation method and transmission apparatus for transmitting parity data for error correction suitable for IP transmission.

  In recent years, services for transmitting video / audio data using Internet protocol (IP) via various communication media such as Ethernet (registered trademark) have become widespread. In particular, stream-type transmission services such as video on demand (VOD) have been provided.

  In the stream transmission service, a transmitter first packetizes video / audio data into a predetermined format and transmits the packet on a network. The receiver receives the streaming packet on the network, extracts the video / audio data included in the packet, transmits it to the decoder, and reproduces the video / audio.

  Hereinafter, the transmitted data on the network is referred to as a streaming packet.

  In general, data transmission is not guaranteed for data transmission over the Internet. When packet loss occurs on the network, the received video / audio data is lost, so that the video / audio is disturbed.

  Packet loss often occurs due to an overflow of a buffer for storing a packet queue in a repeater such as a router constituting a network. For this reason, when performing burst transfer in which streaming packets are transmitted in a certain size at the time of transmission, there is a tendency that the later the packet that is burst transferred, the higher the probability of buffer overflow and the more likely packet loss occurs.

  In order to solve this problem, there is an FEC (Forward Error Correction) method in which a redundant packet for error correction is transmitted together with a streaming packet on the transmitter side, and data lost by the redundant packet is restored when data is lost on the receiving side. Proposed. Among them, a proposal has been made to reduce the packet loss as much as possible and suppress the occurrence of packet loss (see Patent Document 1).

  Hereinafter, conventional examples will be described with reference to the drawings. FIG. 2 shows a configuration of a conventional transmission apparatus. The encoder 22 compresses the video data and stores it in the buffer 23. The RTP packet generation unit 24 reads the compressed video data in a predetermined unit, adds an IP header 42, a UDP header 43, and an RTP header 44 as shown in FIG. 4 to transmit to the communication unit 25, and transmits the packet to the network. The data is transmitted and stored in the packet buffer 28.

  The FEC packet generation unit 26 generates FEC packets from the RTP packets accumulated in the packet buffer 28 and stores them in the packet buffer 27. The stored FEC packet is transmitted to the communication unit 25 and output as a predetermined packet.

  FIG. 3 shows a stream packet on the transmission line in this conventional embodiment. The conventional example shows a case where four packets are stored in the packet buffer. After four RTP packets are created, they are output together on the network via the communication unit 25.

  After outputting the RTP packet, the FEC packet generation unit 26 creates a parity packet from the RTP packet stored in the packet buffer 28.

  The RTP packet generation unit 24, after outputting the packet, calls the video data from the buffer 23 at a predetermined timing, creates an RTP packet (4 packets in the conventional example), and outputs it to the communication unit 25. When there is FEC parity data of the RTP packet output last time in the generation unit, the parity packet for FEC is output, and the packet stored in the packet buffer 28 is subsequently output to the network. Such an operation is repeated, and as shown in FIG. 3, the parity packet 32 and the RTP packet 31 become a lump, and the packet group is transferred at a predetermined interval.

  A feature of this conventional method is that the RTP packet generator reads data from the buffer 23 at a predetermined interval, so that more than five packets including the FEC parity packet are not collectively output to the network. In this way, a method has been proposed in which a predetermined number of packet groups are transmitted to a network at regular intervals to reduce the probability of packet loss.

  Since various devices are connected to a network represented by Ethernet (registered trademark), even if stream packets to be transmitted are transmitted at regular intervals as in the conventional example, other packets are not transmitted. This may cause the repeater buffer to overflow. In particular, since the performance improvement of personal computers in recent years is remarkable, if a high-speed file copy or the like is performed between PCs during transmission of streaming packets, the number of packets input to the repeater increases and the probability of buffer overflow increases.

  In the repeater, when the buffer overflows, subsequent packets are likely to be lost collectively.

  In the case of the conventional example, when a maximum of 5 packets are transmitted together, the buffer of the repeater is already full due to a large number of packets such as file copy between PCs before the streaming packet is received. In this case, five streaming packets are lost together.

  In the case of the conventional embodiment, one parity packet is assigned to four packets. However, with such a parity assignment method, it is generally possible to reproduce lost packets of one of five packets including parity. However, when two or more packets are lost, the lost packet cannot be reproduced.

  In order to enable correction even when packet losses occur collectively, a highly efficient correction code such as interleaved Reed-Solomon is often used. An example of the RTP packet format by the interleaved Reed-Solomon code is shown in FIG.

  In the conventional embodiment, 8 packets of parity packet group 52 are added to 32 packets of data packet group 51. The redundancy is 20% because 8 out of 40 packets are parity packets, which is the same as the conventional system in which one parity packet is provided for 4 packets. When the 8-packet Reed-Solomon code of the conventional embodiment is used, correction is possible even if there is an 8-packet loss.

  In this way, by collecting data in large units and generating a parity code, even if there are a plurality of packet losses, correction can be performed efficiently.

  However, in the case of realizing the conventional method as described in Patent Document 1, first, the packet buffer 28 has a problem that a large capacity is required because video data is buffered in units of error correction parity. For example, in the case of the conventional example, what is normally 4 packets × 1500B = 6000B is 32 packets × 1500B = 48000B, which requires 8 times as many buffers.

  In the conventional embodiment, since the RTP packet composed of video data and the parity packet of the RTP packet sent last time are sent together, in the case of a high-efficiency encoding method such as interleaved Reed-Solomon Since the burst length of the outgoing packet is long (40 packets in this description), there is a problem that packets flowing on the transmission path are not dispersed and the probability of occurrence of packet loss increases.

  Further, generation of FEC packets is started after the video data is stored in the packet buffer 28 in the RTP packet generation unit 24. However, in the conventional embodiment, since it is necessary to generate parity collectively, 1 There is a problem that the amount of calculation increases. In the case of the conventional example, since the next video data cannot be sent out until the generation of the parity packet is completed in the FEC packet generator 26, in order to transmit high-quality data such as a high-definition video, the FEC The packet generation unit 26 needs to be realized by hardware.

  In this method, a large-capacity buffer is required to generate a parity code after data is temporarily stored. In order to realize the FEC packet generation unit 26 at a low cost by hardware, a buffer for video data is required. Memory selection is important.

  In general, an SD-RAM with a low unit price per bit has a high speed for continuous address reading but a slow reading of jumping addresses. The FEC parity generation circuit must access the data in the order indicated by the arrow 53 in FIG. 5 to generate parity data, so the memory address of the buffer is accessed in a jump, and SD-RAM is used as the buffer. In some cases, the calculation time is extremely long. For this reason, when high-definition video data is transmitted, an expensive large-capacity SRAM with fast random access is required, and it is difficult to realize an FEC generation circuit at a low cost.

In the transmission apparatus described in Patent Document 1, the RTP packet header addition of video data is performed by the RTP packet generation unit 24, and the RTP header addition of FEC parity data is performed by the FEC packet generation unit 26. Due to the complexity, there is a problem that the circuit scale becomes large when hardware is used.
JP 2005-136546 A

  In view of the above problems, the present invention can be realized at high speed with a small amount of memory even when an efficient error correction code that generates an error correction code in a large unit such as interleaved Reed-Solomon is used, and on the network. It is an object of the present invention to provide an FEC packet generation method and a transmission apparatus that can reduce the probability of occurrence of packet loss and perform highly reliable data transmission by distributing packets evenly.

  In order to achieve the above object, the present invention provides parity generation means for generating K-byte parity data in units of N bytes for input data, and the input data or the parity data generated by the parity generation means. Selection means for selecting and outputting one of them, division means for dividing the input data or the parity data into units of M bytes, and adding a header to the input data or the parity data divided by the division means RTP packet generation means for outputting an RTP packet, wherein N and K are each an integral multiple of M, and the selection means transmits the input data to the RTP packet generation means as N bytes. After transmitting the K-byte parity data, the RTP packet generation means Storage means for storing the generated RTP packets, timer means for generating timing to output to the network at predetermined intervals, and communication means for outputting the RTP packets to the network in accordance with the timing generated by the timer generating means It was set as the structure provided with.

  According to the present invention, N and K are each an integral multiple of M, and the selection unit transmits K bytes of parity data to the RTP packet generation unit after transmitting N bytes of input data, and the RTP Accumulating means for storing the RTP packet generated by the packet generating means, timer means for generating timing to output to the network at a predetermined interval, and sending the RTP packet to the network according to the timing generated by the timer generating means Since it has a communication means for output, it is possible to generate a highly efficient error correction code with a simple configuration and to transmit video data to the network without increasing the probability of packet loss. A transmission apparatus can be provided.

  According to the present invention, N and K are each an integral multiple of M, and the selection unit transmits K bytes of parity data to the RTP packet generation unit after transmitting N bytes of input data, and the RTP Accumulating means for storing the RTP packet generated by the packet generating means, timer means for generating timing to output to the network at a predetermined interval, and sending the RTP packet to the network according to the timing generated by the timer generating means Communication means for outputting.

  As a result, it is possible to generate a highly efficient error correction code with a simple configuration, and to transmit video data to the network without increasing the probability of occurrence of packet loss. Is possible.

  The parity generation means may have a K-byte memory.

  As a result, the capacity necessary and sufficient for parity generation can be obtained, and the circuit configuration can be reduced.

  Further, the parity generation means may be configured to generate parity by interleaved Reed-Solomon.

  As a result, a more efficient error correction code can be generated.

  The parity generation means includes an RS encoder composed of a plurality of registers and a parity memory, sets the contents of the parity memory in the plurality of registers, and inputs one symbol data from the input data. A configuration may be adopted in which parity calculation is performed by an RS encoder, the contents of the register after the parity calculation are stored in the same memory address read from the parity memory, and a memory address for performing the next input data processing is designated. .

  As a result, the parity can be generated with a small amount of memory without accumulating data, so that data can be transmitted at high speed.

  In addition, when the throughput of the input data is X (bytes / second), the setting value of the timer unit may be set to be equal to or less than the time obtained by (M / X × N / (N + K)).

  As a result, the timer can be set to a value that shortens the packet transmission interval by the amount that the parity packet is added, and packets can be transmitted to the network at approximately equal intervals, which is as efficient as a Reed-Solomon code. Even when the error correction code is used, video data can be transmitted without increasing the error occurrence probability.

  Further, when the input data is real-time video, the data may be input via a FIFO memory.

  As a result, it is possible to eliminate data loss that occurs when reception of a real-time video transmitted in synchronization with an external clock is stopped.

(Embodiment)
Embodiments of an FEC parity generation method and a transmission apparatus according to the present invention will be described below.

  FIG. 1 is a diagram illustrating an example of a configuration of a transmission apparatus according to the present invention. The transmission apparatus 100 of the present invention comprises a FIFO 13, a selection unit 14, a parity generation unit 15, an MTU division unit 16, a second selection unit 17, an RTP packet generation unit 18, a storage unit 19, a communication unit 20, and a timer unit 21. .

  The real-time video data 11 is an output signal from a digital tuner, an MPEG encoder, or the like, and is input to the transmission device 100 according to the present invention.

  The non-real-time video data 12 is video data recorded on an HDD, a DVD, or the like, and is called as necessary and input to the transmission device 100 of the present invention.

  The FIFO 13 is provided for the purpose of preventing video data from being lost when the processing load of the transmission apparatus 100 is high and real-time video data cannot be received.

  The selection means 14 selects either real-time video data or non-real-time video data.

  The parity generation unit 15 receives the video data selected by the selection unit 14 and generates a parity that can be decoded by the receiving device when a packet is lost during transfer through the network. The parity generation unit 15 generates parity in units of N bytes for the input video data, and transmits the parity data to the second selection unit 17 at the subsequent stage.

  The MTU dividing unit 16 divides the input data size into M bytes having a predetermined length and outputs it to the network, and generates parity generation and timing for outputting K-byte parity data to the network.

  The RTP packet generation unit 18 adds a header necessary for transmission to the network to the input M-byte data and transfers the data to the storage unit 19.

  The timer means 21 notifies a timing signal to the communication means every set time.

  The communication unit 20 reads out data from the storage unit 19 using the timing signal as an activation trigger, and transmits the stream packet to the network.

  The operation of the transmission apparatus 100 configured as described above will be described below.

  First, a case where non-real time video data is selected by the selection unit 14 will be described.

  First, as preparation for starting data transfer, a counter (not shown) in the MTU dividing unit 16 is initialized, and start timing information is transmitted to the parity generating unit 15 and the RTP packet generating unit 18.

  When the parity generation means 15 receives the start timing signal, it initializes the contents of the internal registers and memory.

  The RTP packet generator 18 enters a reception standby state.

  When the MTU dividing unit 16 is initialized, the second selecting unit 17 is set to transmit the video data output from the selecting unit 14 to the RTP packet generating unit 18.

  When initialization is completed as described above, non-real-time video data is called for each byte, passes through the second selection means 17 and is transmitted to the RTP packet generation means 18, and M bytes are transmitted to the RTP packet generation means 18. When accumulated, as shown in FIG. 4, an RTP header 44, a UDP header 43, and an IP header 42 are added to the M-byte data 41 and sequentially transferred to the accumulation means 19.

  The parity generation means 15 generates a K-byte parity code for error correction from the input data. Although there are various methods for error correction codes, Reed-Solomon codes are often used. FIG. 5 shows a data format including an error correction code.

  In this embodiment, MTU (Maximum Transmission Unit) is 1500 B (bytes) as an example, and M = 1454 bytes, K = 1454 × 8 = 111632 bytes, and N = 1454 × 32 = 46528 bytes.

  The parity generation unit 15 generates a K-byte parity packet for 8 packets from N-byte data for 32 packets.

  In general parity generation methods, interleave processing is performed, so that N bytes of data are temporarily stored in the parity generation means 15, and the head data of each packet is converted into an RS encoder as indicated by an arrow 53 in FIG. (Sequentially input) to generate 8-byte parity data, then input the second byte data of each packet to the RS encoder to generate parity, and repeat this for 1454 bytes of data length Thus, parity data for the parity packet group 52 is created.

  When the parity packet is generated, the second selection unit 17 is switched to transmit the parity packet to the RTP packet generation unit 18, and the IP header, the UDP header, and the RTP header are added to the network to be transmitted to the network.

  In the case of the present embodiment, since 8 error correction packets are added to 32 data packets, parity packets are used even when 8 packet losses occur in total 40 packets. The performance can be restored.

  In order to create such a parity packet, a 46528B buffer memory is required, but this can be configured with a smaller capacity, and a configuration in which performance is not deteriorated even with a memory with a slow random access such as SD-RAM. Next, it will be described.

  In FIG. 6, input data is input to the RS encoder 61. The RS encoder 61 is connected to the memory means 62. An address counter 63 is connected to the memory means 62 and counts the number of input data. When generating parity as shown in FIG. 5, data is counted cyclically from 1 to 32.

  FIG. 7 shows the configuration of the RS encoder 61. In the RS encoder 61, the multiplier 72 is a Galois multiplier, and multiplication coefficients A to H based on Reed-Solomon code theory are set. The adder 73 means addition in the Galois field, and an EX-OR (exclusive OR) operation is actually performed.

  In normal parity operation, every time data is input in order, the operation and the shift of the register are performed, and the last remaining register value becomes the parity data, but this embodiment is different from this. It is configured to be connected to the memory bus so that the preset can be performed, and the parity is generated by the following procedure.

  Before the data is input, the memory contents of the memory means 62 are set in the register.

  After the contents of the memory are set in the register, the value of the register after one byte of data is input and the operation and shift are performed is written back to the memory again.

  After writing back to the memory, the address counter 63 is counted and the same processing is repeated.

  In the case of this embodiment, since the address circulates with 32 pieces of input data, after the data of 32 packets of 1454 bytes have passed, the parity 1 to 8 of the parity packet group 52 shown in FIG. Will be stored.

  As described above, according to the configuration of the parity generation unit 15 as shown in FIG. 6, it is not necessary to store data, and it is only necessary to provide a buffer memory for parity, so that parity can be generated with a small amount of memory. In addition, since the generation of parity is completed when N bytes have passed, there is no need to store all the data once and store the data as in the conventional method, so that data can be transmitted at high speed. Become.

  Thus, when N bytes of data pass through the RTP packet generation unit 18, the parity generation unit 15 has already generated parity, and the second selection unit 17 stores the parity stored in the memory in the parity generation unit 15. The data is transferred to the RTP packet generation means 18, added with an RTP header, and output to the storage means 19.

  In the case of this embodiment, after the parity packet of 8 packets is transferred to the storage means 19, the parity generation means 15 is initialized, and thereafter the same operation is repeated.

  The timer unit 21 is set with a time interval for outputting the RTP packets stored in the storage unit 19. This time is a value obtained by (Expression 1), assuming that the transfer rate for outputting the non-real-time video data 12 is X bytes / second.

  In this way, by setting the timer with a value that shortens the packet transmission interval by the amount that the parity packet is added, packets are transmitted to the network at approximately equal intervals. Even when a highly efficient error correction code is used, video data can be transmitted without increasing the error occurrence probability.

  Further, the storage means 19 can be configured with a small amount of memory because it is sufficient to provide only a buffer that can store the number of parity packets at the maximum. In this embodiment, it can be composed of K bytes.

  Next, the case of real-time video will be described.

  When the real-time video data 11 is input, the data transmission apparatus according to the embodiment of the present invention transfers the K-byte parity data stored in the parity generation unit 15 to the RTP packet generation unit 18 after the N-byte data transfer. In the meantime, no parity can be generated. Since the real-time video data 11 is transmitted in synchronization with an external clock, data loss occurs when the reception is stopped. The FIFO memory 13 is provided.

  With this configuration, even a highly efficient code such as a Reed-Solomon code can generate parity at high speed with a small amount of memory, and the RTP packet generation means can be used by switching between normal data and parity. Therefore, it can be realized with a simple configuration.

  Since RTP packets can be temporarily stored and output to the network at a predetermined time interval set in the timer means, even when an error correction code having a plurality of parity packets is used, the RTP packets are evenly distributed to the network. Since it is output at intervals, it is possible to transmit a parity packet without increasing the error occurrence probability, and therefore, it is possible to provide a high-quality and low-cost transmission device.

  In the embodiment of the present invention, eight parity packets are used for 32 data packets, but this is not fixed. For example, if 64 data packets are replaced with 16 parity packets, data can be reproduced with the same redundancy even if 16 of 80 packets are lost.

  In the embodiment of the present invention, the Reed-Solomon code has been described. However, since the effect of the present invention is an effective means for adding K-byte parity data to N-byte data, error correction is performed. The encoding method is not limited to Reed Solomon.

  In the embodiment of the present invention, the case where the MTU is 1500B has been described as an example. However, other MTUs may be used.

  The present invention relates to a transmission apparatus and method, and in particular, can be applied to a transmission apparatus in which error correction data can be added and transmitted.

Configuration diagram of transmitting apparatus of the present invention Configuration diagram of conventional transmitter The figure which shows the packet on the network when transmitting with the conventional transmission equipment Transmission data configuration diagram Illustration of error correction code by interleaved Reed-Solomon Configuration diagram of parity generation means in the present invention Configuration diagram of an RS encoder in the present invention

Explanation of symbols

11 real-time video data 12 non-real-time video data 13 FIFO memory 14 selection means 15 parity generation means 16 MTU division means 17 second selection means 18 RTP packet generation means 19 storage means 20 communication means 21 timer means 22 encoder 23 buffer 24 RTP packet Generation unit 25 Communication unit 26 FEC packet generation unit 27 Packet buffer 28 Packet buffer 31 Transmission packet group 32 Parity packet 41 Data packet 42 IP header 43 UDP header 44 RTP header 51 Data packet group 52 Parity packet group 61 RS encoder 62 Memory means 63 Address counter 72 Galois multiplier 73 Adder 100 Transmitter

Claims (10)

Parity generation means for generating K-byte parity data in units of N bytes for input data;
Selecting means for selecting and outputting either the input data or the parity data generated by the parity generating means;
A dividing means for dividing the input data or the parity data into units of M bytes;
RTP packet generating means for adding a header to the input data or the parity data divided by the dividing means and outputting an RTP packet;
N and K are each an integer multiple of M;
The selection unit transmits the parity data of K bytes to the RTP packet generation unit after transmitting the input data N bytes,
Storage means for storing RTP packets generated by the RTP packet generation means;
Timer means for generating timing to output to the network at a predetermined interval;
A transmission device comprising: communication means for outputting the RTP packet to a network in accordance with the timing generated by the timer generation means. 2. The transmission apparatus according to claim 1, wherein the parity generation unit includes a K-byte memory. 2. The transmission apparatus according to claim 1, wherein the parity generation unit generates parity by interleaved Reed-Solomon. The parity generation means includes an RS encoder including a plurality of registers, and a parity memory.
Set the contents of the parity memory in the plurality of registers,
1 symbol data is input from the input data and parity calculation is performed by the RS encoder,
Store the contents of the register after the parity operation at the same memory address read from the parity memory;
2. The transmission apparatus according to claim 1, wherein a memory address for performing next input data processing is designated. The set value of the timer means is set to be equal to or less than a time obtained by (M / X × N / (N + K)) where the throughput of the input data is X (bytes / second). 1. The transmission device according to 1. 2. The transmission apparatus according to claim 1, wherein the storage means has a capacity equivalent to a capacity of a parity memory in the parity generation means. 2. The transmission apparatus according to claim 1, wherein when the input data is a real-time video, data is input via a FIFO memory. Generate K bytes of parity data in units of N bytes for the input data,
Selectively output either the input data or the parity data,
In the parity generation method for FEC that divides the input data or the parity data into units of M bytes, adds a header, and outputs an RTP packet,
N and K are each an integer multiple of M;
The selection output is to transmit the parity data of the K bytes after transmitting the input data N bytes,
Generate the timing to output to the network at a predetermined interval,
A parity generation method for FEC, wherein the RTP packet is output to a network according to the timing. 9. The parity generation method for FEC according to claim 8, wherein the generation of the parity data is interleaved Reed-Solomon. 9. The parity generation method for FEC according to claim 8, wherein when the throughput of the input data is X (bytes / second), the timing is equal to or less than a time obtained by (M / X × N / (N + K)). .

Images