1.1. Limits of Block Codes with Real-Time Flows

With FECFRAME, there is a single FEC encoding point (either an end host/server (source) or a middlebox) and a single FEC decoding point per receiver (either an end host (receiver) or middlebox). In this context, currently standardized AL-FEC codes for FECFRAME like Reed-Solomon [RFC6865], LDPC-Staircase [RFC6816], or Raptor/RaptorQ [RFC6681], are all linear block codes: they require the data flow to be segmented into blocks of a predefined maximum size.¶

To define this block size, it is required to find an appropriate balance between robustness and decoding latency: the larger the block size, the higher the robustness (e.g., in case of long packet erasure bursts), but also the higher the maximum decoding latency (i.e., the maximum time required to recover a lost (erased) packet thanks to FEC protection). Therefore, with a multicast/broadcast session where different receivers experience different packet loss rates, the block size should be chosen by considering the worst communication conditions one wants to support, but without exceeding the desired maximum decoding latency. This choice then impacts the FEC-related latency of all receivers, even those experiencing a good communication quality, since no FEC encoding can happen until all the source data of the block is available at the sender, which directly depends on the block size.¶

1.2. Lower Latency and Better Protection of Real-Time Flows with the Sliding Window RLC Codes

This document introduces two fully specified FEC schemes that do not follow the block code approach: the Sliding Window Random Linear Codes (RLC) over either Galois Fields (a.k.a., Finite Fields) GF(2) (the "binary case") or GF(2⁸), each time with the possibility of controlling the code density. These FEC schemes are used to protect arbitrary media streams along the lines defined by FECFRAME extended to sliding window FEC codes [RFC8680]. These FEC schemes and, more generally, Sliding Window FEC codes are recommended, for instance, with media that feature real-time constraints sent within a multicast/broadcast session [Roca17].¶

The RLC codes belong to the broad class of sliding-window AL-FEC codes (a.k.a., convolutional codes) [RFC8406]. The encoding process is based on an encoding window that slides over the set of source packets (in fact source symbols as we will see in Section 3.2), this window being either of fixed size or variable size (a.k.a., an elastic window). Repair symbols are generated on-the-fly, by computing a random linear combination of the source symbols present in the current encoding window, and passed to the transport layer.¶

At the receiver, a linear system is managed from the set of received source and repair packets. New variables (representing source symbols) and equations (representing the linear combination carried by each repair symbol received) are added upon receiving new packets. Variables and the equations they are involved in are removed when they are too old with respect to their validity period (real-time constraints). Lost source symbols are then recovered thanks to this linear system whenever its rank permits to solve it (at least partially).¶

The protection of any multicast/broadcast session needs to be dimensioned by considering the worst communication conditions one wants to support. This is also true with RLC (more generally, any sliding window) code. However, the receivers experiencing a good to medium communication quality will observe a reduced FEC-related latency compared to block codes [Roca17] since an isolated lost source packet is quickly recovered with the following repair packet. On the opposite, with a block code, recovering an isolated lost source packet always requires waiting for the first repair packet to arrive after the end of the block. Additionally, under certain situations (e.g., with a limited FEC-related latency budget and with constant bitrate transmissions after FECFRAME encoding), sliding window codes can more efficiently achieve a target transmission quality (e.g., measured by the residual loss after FEC decoding) by sending fewer repair packets (i.e., higher code rate) than block codes.¶

1.3. Small Transmission Overheads with the Sliding Window RLC FEC Scheme

The Sliding Window RLC FEC scheme is designed to limit the packet header overhead. The main requirement is that each repair packet header must enable a receiver to reconstruct the set of source symbols plus the associated coefficients used during the encoding process. In order to minimize packet overhead, the set of source symbols in the encoding window as well as the set of coefficients over GF(2^m) (where m is 1 or 8, depending on the FEC scheme) used in the linear combination are not individually listed in the repair packet header. Instead, each FEC Repair Packet header contains:¶

• the Encoding Symbol Identifier (ESI) of the first source symbol in the encoding window as well as the number of symbols (since this number may vary with a variable size, elastic window). These two pieces of information enable each receiver to reconstruct the set of source symbols considered during encoding, the only constraint being that there cannot be any gap;¶
• the seed and density threshold parameters used by a coding coefficients generation function (Section 3.6). These two pieces of information enable each receiver to generate the same set of coding coefficients over GF(2^m) as the sender;¶

Therefore, no matter the number of source symbols present in the encoding window, each FEC Repair Packet features a fixed 64-bit long header, called Repair FEC Payload ID (Figure 8). Similarly, each FEC Source Packet features a fixed 32-bit long trailer, called Explicit Source FEC Payload ID (Figure 6), that contains the ESI of the first source symbol (Section 3.2).¶

3.1. Codec Parameters

A codec implementing the Sliding Window RLC FEC scheme relies on several parameters:¶

Maximum FEC-related latency budget, max_lat (a decimal number expressed in seconds) with real-time flows:

a source ADU flow can have real-time constraints, and therefore any FECFRAME related operation should take place within the validity period of each ADU (Appendix D describes an exception to this rule). When there are multiple flows with different real-time constraints, we consider the most stringent constraints (see item 6 in Section 10.2 of [RFC6363], for recommendations when several flows are globally protected). The maximum FEC-related latency budget, max_lat, accounts for all sources of latency added by FEC encoding (at a sender) and FEC decoding (at a receiver). Other sources of latency (e.g., added by network communications) are out of scope and must be considered separately (said differently, they have already been deducted from max_lat). max_lat can be regarded as the latency budget permitted for all FEC-related operations. This is an input parameter that enables a FECFRAME sender to derive other internal parameters (see Appendix C);¶

Encoding window current (resp. maximum) size, ew_size (resp. ew_max_size) (in symbols):

at a FECFRAME sender, during FEC encoding, a repair symbol is computed as a linear combination of the ew_size source symbols present in the encoding window. The ew_max_size is the maximum size of this window, while ew_size is the current size. For example, in the common case at session start, upon receiving new source ADUs, the ew_size progressively increases until it reaches its maximum value, ew_max_size. We have:¶

• 0 < ew_size <= ew_max_size¶

Decoding window maximum size, dw_max_size (in symbols):

at a FECFRAME receiver, dw_max_size is the maximum number of received or lost source symbols that are still within their latency budget;¶

Linear system maximum size, ls_max_size (in symbols):

at a FECFRAME receiver, the linear system maximum size, ls_max_size, is the maximum number of received or lost source symbols in the linear system (i.e., the variables). It SHOULD NOT be smaller than dw_max_size since it would mean that, even after receiving a sufficient number of FEC Repair Packets, a lost ADU may not be recovered just because the associated source symbols have been prematurely removed from the linear system, which is usually counter-productive. On the opposite, the linear system MAY grow beyond the dw_max_size (Appendix D);¶

Symbol size, E (in bytes):

the E parameter determines the source and repair symbol sizes (necessarily equal). This is an input parameter that enables a FECFRAME sender to derive other internal parameters, as explained below. An implementation at a sender MUST fix the E parameter and MUST communicate it as part of the FEC Scheme-Specific Information (Section 4.1.1.2).¶

Code rate, cr:

The code rate parameter determines the amount of redundancy added to the flow. More precisely the cr is the ratio between the total number of source symbols and the total number of source plus repair symbols and by definition: 0 < cr <= 1. This is an input parameter that enables a FECFRAME sender to derive other internal parameters, as explained below. However, there is no need to communicate the cr parameter per see (it's not required to process a repair symbol at a receiver). This code rate parameter can be static. However, in specific use-cases (e.g., with unicast transmissions in presence of a feedback mechanism that estimates the communication quality, out of scope of FECFRAME), the code rate may be adjusted dynamically.¶

Appendix C proposes non-normative techniques to derive those parameters, depending on the use-case specificities.¶

3.2. ADU, ADUI, and Source Symbols Mappings

At a sender, an ADU coming from the application is not directly mapped to source symbols. When multiple source flows (e.g., media streams) are mapped onto the same FECFRAME instance, each flow is assigned its own Flow ID value (see below). This Flow ID is then prepended to each ADU before FEC encoding. This way, FEC decoding at a receiver also recovers this Flow ID and the recovered ADU can be assigned to the right source flow (note that the 5-tuple used to identify the right source flow of a received ADU is absent with a recovered ADU since it is not FEC protected).¶

Additionally, since ADUs are of variable size, padding is needed so that each ADU (with its flow identifier) contribute to an integral number of source symbols. This requires adding the original ADU length to each ADU before doing FEC encoding. Because of these requirements, an intermediate format, the ADUI, or ADU Information, is considered [RFC6363].¶

For each incoming ADU, an ADUI MUST be created as follows. First of all, 3 bytes are prepended (Figure 1):¶

Flow ID (F) (8-bit field):

this unsigned byte contains the integer identifier associated to the source ADU flow to which this ADU belongs. It is assumed that a single byte is sufficient, which implies that no more than 256 flows will be protected by a single FECFRAME session instance.¶

Length (L) (16-bit field):

this unsigned integer contains the length of this ADU, in network byte order (i.e., big endian). This length is for the ADU itself and does not include the F, L, or Pad fields.¶

Then, zero padding is added to the ADU if needed:¶

Padding (Pad) (variable size field):

this field contains zero padding to align the F, L, ADU and padding up to a size that is multiple of E bytes (i.e., the source and repair symbol length).¶

The data unit resulting from the ADU and the F, L, and Pad fields is called ADUI. Since ADUs can have different sizes, this is also the case for ADUIs. However, an ADUI always contributes to an integral number of source symbols.¶

   symbol length, E              E                     E
< ------------------ >< ------------------ >< ------------------ >
+-+--+---------------------------------------------+-------------+
|F| L|                     ADU                     |     Pad     |
+-+--+---------------------------------------------+-------------+

Here, three source symbols are created for this ADUI.¶

Note that neither the initial 3 bytes nor the optional padding are sent over the network. However, they are considered during FEC encoding, and a receiver that lost a certain FEC Source Packet (e.g., the UDP datagram containing this FEC Source Packet when UDP is used as the transport protocol) will be able to recover the ADUI if FEC decoding succeeds. Thanks to the initial 3 bytes, this receiver will get rid of the padding (if any) and identify the corresponding ADU flow.¶

3.3. Encoding Window Management

Source symbols and the corresponding ADUs are removed from the encoding window:¶

• when the sliding encoding window has reached its maximum size, ew_max_size. In that case the oldest symbol MUST be removed before adding a new symbol, so that the current encoding window size always remains inferior or equal to the maximum size: ew_size <= ew_max_size;¶
• when an ADU has reached its maximum validity duration in case of a real-time flow. When this happens, all source symbols corresponding to the ADUI that expired SHOULD be removed from the encoding window;¶

Source symbols are added to the sliding encoding window each time a new ADU arrives, once the ADU-to-source symbols mapping has been performed (Section 3.2). The current size of the encoding window, ew_size, is updated after adding new source symbols. This process may require to remove old source symbols so that: ew_size <= ew_max_size.¶

Note that a FEC codec may feature practical limits in the number of source symbols in the encoding window (e.g., for computational complexity reasons). This factor may further limit the ew_max_size value, in addition to the maximum FEC-related latency budget (Section 3.1).¶

3.4. Source Symbol Identification

Each source symbol is identified by an Encoding Symbol ID (ESI), an unsigned integer. The ESI of source symbols MUST start with value 0 for the first source symbol and MUST be managed sequentially. Wrapping to zero happens after reaching the maximum value made possible by the ESI field size (this maximum value is FEC scheme dependent, for instance, 2³²-1 with FEC schemes 9 and 10).¶

No such consideration applies to repair symbols.¶

3.5. Pseudorandom Number Generator (PRNG)

In order to compute coding coefficients (see Section 3.6), the RLC FEC schemes rely on the TinyMT32 PRNG defined in [RFC8682] with two additional functions defined in this section.¶

This PRNG MUST first be initialized with a 32-bit unsigned integer, used as a seed, with:¶

   void   tinymt32_init (tinymt32_t * s, uint32_t seed);

With the FEC schemes defined in this document, the seed is in practice restricted to a value between 0 and 0xFFFF inclusive (note that this PRNG accepts a seed value equal to 0), since this is the Repair_Key 16-bit field value of the Repair FEC Payload ID (Section 4.1.3). In practice, how to manage the seed and Repair_Key values (both are equal) is left to the implementer, using a monotonically increasing counter being one possibility (Section 6.1). In addition to the seed, this function takes as parameter a pointer to an instance of a tinymt32_t structure that is used to keep the internal state of the PRNG.¶

Then, each time a new pseudorandom integer between 0 and 15 inclusive (4-bit pseudorandom integer) is needed, the following function is used:¶

   uint32_t   tinymt32_rand16 (tinymt32_t * s);

This function takes as parameter a pointer to the same tinymt32_t structure (that is left unchanged between successive calls to the function).¶

Similarly, each time a new pseudorandom integer between 0 and 255 inclusive (8-bit pseudorandom integer) is needed, the following function is used:¶

   uint32_t   tinymt32_rand256 (tinymt32_t * s);
 

These two functions keep respectively the 4 or 8 less significant bits of the 32-bit pseudorandom number generated by the tinymt32_generate_uint32() function of [RFC8682]. This is done by computing the result of a binary AND between the tinymt32_generate_uint32() output and respectively the 0xF or 0xFF constants, using 32-bit unsigned integer operations. Figure 2 shows a possible implementation. This is a C language implementation, written for C99 [C99]. Test results discussed in Appendix B show that this simple technique, applied to this PRNG, is in line with the RLC FEC schemes needs.¶

<CODE BEGINS> file ""

/**
 * This function outputs a pseudorandom integer in [0 .. 15] range.
 *
 * @param s pointer to tinymt internal state.
 * @return unsigned integer between 0 and 15 inclusive.
 */
uint32_t tinymt32_rand16(tinymt32_t *s)
{
    return (tinymt32_generate_uint32(s) & 0xF);
}

/**
 * This function outputs a pseudorandom integer in [0 .. 255] range.
 *
 * @param s pointer to tinymt internal state.
 * @return unsigned integer between 0 and 255 inclusive.
 */
uint32_t tinymt32_rand256(tinymt32_t *s)
{
    return (tinymt32_generate_uint32(s) & 0xFF);
}

<CODE ENDS>

Any implementation of this PRNG MUST have the same output as that provided by the reference implementation of [RFC8682]. In order to increase the compliancy confidence, three criteria are proposed: the one described in [RFC8682] (for the TinyMT32 32-bit unsigned integer generator), and the two others detailed in Appendix A (for the mapping to 4-bit and 8-bit intervals). Because of the way the mapping functions work, it is unlikely that an implementation that fulfills the first criterion fails to fulfill the two others.¶

3.6. Coding Coefficients Generation Function

The coding coefficients, used during the encoding process, are generated at the RLC encoder by the generate_coding_coefficients() function each time a new repair symbol needs to be produced. The fraction of coefficients that are nonzero (i.e., the density) is controlled by the DT (Density Threshold) parameter. DT has values between 0 (the minimum value) and 15 (the maximum value), and the average probability of having a nonzero coefficient equals (DT + 1) / 16. In particular, when DT equals 15 the function guaranties that all coefficients are nonzero (i.e., maximum density).¶

These considerations apply to both the RLC over GF(2) and RLC over GF(2⁸), the only difference being the value of the m parameter. With the RLC over GF(2) FEC scheme (Section 5), m is equal to 1. With RLC over GF(2⁸) FEC scheme (Section 4), m is equal to 8.¶

Figure 3 shows the reference generate_coding_coefficients() implementation. This is a C language implementation, written for C99 [C99].¶

<CODE BEGINS> file ""

#include <string.h>

/*
 * Fills in the table of coding coefficients (of the right size)
 * provided with the appropriate number of coding coefficients to
 * use for the repair symbol key provided.
 *
 * (in) repair_key    key associated to this repair symbol. This
 *                    parameter is ignored (useless) if m=1 and dt=15
 * (in/out) cc_tab    pointer to a table of the right size to store
 *                    coding coefficients. All coefficients are
 *                    stored as bytes, regardless of the m parameter,
 *                    upon return of this function.
 * (in) cc_nb         number of entries in the cc_tab table. This
 *                    value is equal to the current encoding window
 *                    size.
 * (in) dt            integer between 0 and 15 (inclusive) that
 *                    controls the density. With value 15, all
 *                    coefficients are guaranteed to be nonzero
 *                    (i.e., equal to 1 with GF(2) and equal to a
 *                    value in {1,... 255} with GF(2^^8)), otherwise
 *                    a fraction of them will be 0.
 * (in) m             Finite Field GF(2^^m) parameter. In this
 *                    document only values 1 and 8 are considered.
 * (out)              returns 0 in case of success, an error code
 *                    different than 0 otherwise.
 */
int generate_coding_coefficients (uint16_t  repair_key,
                                  uint8_t*  cc_tab,
                                  uint16_t  cc_nb,
                                  uint8_t   dt,
                                  uint8_t   m)
{
    uint32_t      i;
    tinymt32_t    s;    /* PRNG internal state */

    if (dt > 15) {
        return -1; /* error, bad dt parameter */
    }
    switch (m) {
    case 1:
        if (dt == 15) {
            /* all coefficients are 1 */
            memset(cc_tab, 1, cc_nb);
        } else {
            /* here coefficients are either 0 or 1 */
            tinymt32_init(&s, repair_key);
            for (i = 0 ; i < cc_nb ; i++) {
                cc_tab[i] = (tinymt32_rand16(&s) <= dt) ? 1 : 0;
            }
        }
        break;

    case 8:
        tinymt32_init(&s, repair_key);
        if (dt == 15) {
            /* coefficient 0 is avoided here in order to include
             * all the source symbols */
            for (i = 0 ; i < cc_nb ; i++) {
                do {
                    cc_tab[i] = (uint8_t) tinymt32_rand256(&s);
                } while (cc_tab[i] == 0);
            }
        } else {
            /* here a certain number of coefficients should be 0 */
            for (i = 0 ; i < cc_nb ; i++) {
                if (tinymt32_rand16(&s) <= dt) {
                    do {
                        cc_tab[i] = (uint8_t) tinymt32_rand256(&s);
                    } while (cc_tab[i] == 0);
                } else {
                    cc_tab[i] = 0;
                }
            }
        }
        break;

    default:
        return -2; /* error, bad parameter m */
    }
    return 0; /* success */
}

<CODE ENDS>

3.7. Finite Field Operations

   repair[i] = cc_tab[0] * src_0[i] XOR cc_tab[1] * src_1[i] XOR ... 
   XOR cc_tab[ew_size - 1] * src_ew_size_1[i]

4.1. Formats and Codes

   fssi=E:1400,WSR:191

 0                   1                   2
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|   Encoding Symbol Length (E)  |      WSR      |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

+--------------------------------+
|           IP Header            |
+--------------------------------+
|        Transport Header        |
+--------------------------------+
|              ADU               |
+--------------------------------+
| Explicit Source FEC Payload ID |
+--------------------------------+

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                   Encoding Symbol ID (ESI)                    |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

+--------------------------------+
|           IP Header            |
+--------------------------------+
|        Transport Header        |
+--------------------------------+
|     Repair FEC Payload ID      |
+--------------------------------+
|         Repair Symbol          |
+--------------------------------+

 0                   1                   2                   3
 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|       Repair_Key              |  DT   |NSS (# src symb in ew) |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
|                            FSS_ESI                            |
+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+

4.2. Procedures

All the procedures of Section 3 apply to this FEC scheme.¶

6.1. Encoding Side

This section provides a high level description of a Sliding Window RLC encoder.¶

Whenever a new FEC Repair Packet is needed, the RLC encoder instance first gathers the ew_size source symbols currently in the sliding encoding window. Then it chooses a repair key, which can be a monotonically increasing integer value, incremented for each repair symbol up to a maximum value of 65535 (as it is carried within a 16-bit field) after which it loops back to 0. This repair key is communicated to the coefficient generation function (Section 3.6) in order to generate ew_size coding coefficients. Finally, the FECFRAME sender computes the repair symbol as a linear combination of the ew_size source symbols using the ew_size coding coefficients (Section 3.7). When E is small and when there is an incentive to pack several repair symbols within the same FEC Repair Packet, the appropriate number of repair symbols are computed. In that case the repair key for each of them MUST be incremented by 1, keeping the same ew_size source symbols, since only the first repair key will be carried in the Repair FEC Payload ID. The FEC Repair Packet can then be passed to the transport layer for transmission. The source versus repair FEC packet transmission order is out of scope of this document and several approaches exist that are implementation-specific.¶

Other solutions are possible to select a repair key value when a new FEC Repair Packet is needed, for instance, by choosing a random integer between 0 and 65535. However, selecting the same repair key as before (which may happen in case of a random process) is only meaningful if the encoding window has changed, otherwise the same FEC Repair Packet will be generated. In any case, choosing the repair key is entirely at the discretion of the sender, since it is communicated to the receiver(s) in each Repair FEC Payload ID. A receiver should not make any assumption on the way the repair key is managed.¶

6.2. Decoding Side

This section provides a high level description of a Sliding Window RLC decoder.¶

A FECFRAME receiver needs to maintain a linear system whose variables are the received and lost source symbols. Upon receiving a FEC Repair Packet, a receiver first extracts all the repair symbols it contains (in case several repair symbols are packed together). For each repair symbol, when at least one of the corresponding source symbols it protects has been lost, the receiver adds an equation to the linear system (or no equation if this repair packet does not change the linear system rank). This equation of course re-uses the ew_size coding coefficients that are computed by the same coefficient generation function (Section 3.6), using the repair key and encoding window descriptions carried in the Repair FEC Payload ID. Whenever possible (i.e., when a sub-system covering one or more lost source symbols is of full rank), decoding is performed in order to recover lost source symbols. Gaussian elimination is one possible algorithm to solve this linear system. Each time an ADUI can be totally recovered, padding is removed (thanks to the Length field, L, of the ADUI) and the ADU is assigned to the corresponding application flow (thanks to the Flow ID field, F, of the ADUI). This ADU is finally passed to the corresponding upper application. Received FEC Source Packets, containing an ADU, MAY be passed to the application either immediately or after some time to guaranty an ordered delivery to the application. This document does not mandate any approach as this is an operational and management decision.¶

With real-time flows, a lost ADU that is decoded after the maximum latency or an ADU received after this delay has no value to the application. This raises the question of deciding whether or not an ADU is late. This decision MAY be taken within the FECFRAME receiver (e.g., using the decoding window, see Section 3.1) or within the application (e.g., using RTP timestamps within the ADU). Deciding which option to follow and whether or not to pass all ADUs, including those assumed late, to the application are operational decisions that depend on the application and are therefore out of scope of this document. Additionally, Appendix D discusses a backward compatible optimization whereby late source symbols MAY still be used within the FECFRAME receiver in order to improve transmission robustness.¶

C.1. Case of a CBR Real-Time Flow

In the following, we consider a real-time flow with max_lat latency budget. The encoding symbol size, E, is constant. The code rate, cr, is also constant, its value depending on the expected communication loss model (this choice is out of scope of this document).¶

In a first configuration, the source ADU flow bitrate at the input of the FECFRAME sender is fixed and equal to br_in (in bits/s), and this value is known by the FECFRAME sender. It follows that the transmission bitrate at the output of the FECFRAME sender will be higher, depending on the added repair flow overhead. In order to comply with the maximum FEC-related latency budget, we have:¶

• dw_max_size = (max_lat * br_in) / (8 * E)¶

assuming that the encoding and decoding times are negligible with respect to the target max_lat. This is a reasonable assumption in many situations (e.g., see Section 8.1 in case of small window sizes). Otherwise the max_lat parameter should be adjusted in order to avoid the problem. In any case, interoperability will never be compromised by choosing a too large value.¶

In a second configuration, the FECFRAME sender generates a fixed bitrate flow, equal to the CBR communication path bitrate equal to br_out (in bits/s), and this value is known by the FECFRAME sender, as in [Roca17]. The maximum source flow bitrate needs to be such that, with the added repair flow overhead, the total transmission bitrate remains inferior or equal to br_out. We have:¶

• dw_max_size = (max_lat * br_out * cr) / (8 * E)¶

assuming here also that the encoding and decoding times are negligible with respect to the target max_lat.¶

For decoding to be possible within the latency budget, it is required that the encoding window maximum size be smaller than or at most equal to the decoding window maximum size. The ew_max_size is the main parameter at a FECFRAME sender, but its exact value has no impact on the FEC-related latency budget. The ew_max_size parameter is computed as follows:¶

• ew_max_size = dw_max_size * WSR / 255¶

In line with [Roca17], WSR = 191 is considered as a reasonable value (the resulting encoding to decoding window size ratio is then close to 0.75), but other values between 1 and 255 inclusive are possible, depending on the use-case.¶

The dw_max_size is computed by a FECFRAME sender but not explicitly communicated to a FECFRAME receiver. However, a FECFRAME receiver can easily evaluate the ew_max_size by observing the maximum Number of Source Symbols (NSS) value contained in the Repair FEC Payload ID of received FEC Repair Packets (Section 4.1.3). A receiver can then easily compute dw_max_size:¶

• dw_max_size = max_NSS_observed * 255 / WSR¶

A receiver can then choose an appropriate linear system maximum size:¶

• ls_max_size >= dw_max_size¶

It is good practice to use a larger value for ls_max_size as explained in Appendix D, which does not impact maximum latency nor interoperability.¶

In any case, for a given use-case (i.e., for target encoding and decoding devices and desired protection levels in front of communication impairments) and for the computed ew_max_size, dw_max_size and ls_max_size values, it is RECOMMENDED to check that the maximum encoding time and maximum memory requirements at a FECFRAME sender, and maximum decoding time and maximum memory requirements at a FECFRAME receiver, stay within reasonable bounds. When assuming that the encoding and decoding times are negligible with respect to the target max_lat, this should be verified as well, otherwise the max_lat SHOULD be adjusted accordingly.¶

The particular case of session start needs to be managed appropriately since the ew_size, starting at zero, increases each time a new source ADU is received by the FECFRAME sender, until it reaches the ew_max_size value. Therefore, a FECFRAME receiver SHOULD continuously observe the received FEC Repair Packets, since the NSS value carried in the Repair FEC Payload ID will increase too, and adjust its ls_max_size accordingly if need be. With a CBR flow, session start is expected to be the only moment when the encoding window size will increase. Similarly, with a CBR real-time flow, the session end is expected to be the only moment when the encoding window size will progressively decrease. No adjustment of the ls_max_size is required at the FECFRAME receiver in that case.¶

C.2. Other Types of Real-Time Flow

In the following, we consider a real-time source ADU flow with a max_lat latency budget and a variable bitrate (VBR) measured at the entry of the FECFRAME sender. A first approach consists in considering the smallest instantaneous bitrate of the source ADU flow, when this parameter is known, and to reuse the derivation of Appendix C.1. Considering the smallest bitrate means that the encoding and decoding window maximum size estimations are pessimistic: these windows have the smallest size required to enable on-time decoding at a FECFRAME receiver. If the instantaneous bitrate is higher than this smallest bitrate, this approach leads to an encoding window that is unnecessarily small, which reduces robustness in front of long erasure bursts.¶

Another approach consists in using ADU timing information (e.g., using the timestamp field of an RTP packet header, or registering the time upon receiving a new ADU). From the global FEC-related latency budget, the FECFRAME sender can derive a practical maximum latency budget for encoding operations, max_lat_for_encoding. For the FEC schemes specified in this document, this latency budget SHOULD be computed with:¶

• max_lat_for_encoding = max_lat * WSR / 255¶

It follows that any source symbols associated to an ADU that has timed-out with respect to max_lat_for_encoding SHOULD be removed from the encoding window. With this approach there is no pre-determined ew_size value: this value fluctuates over the time according to the instantaneous source ADU flow bitrate. For practical reasons, a FECFRAME sender may still require that ew_size does not increase beyond a maximum value (Appendix C.3).¶

With both approaches, and no matter the choice of the FECFRAME sender, a FECFRAME receiver can still easily evaluate the ew_max_size by observing the maximum Number of Source Symbols (NSS) value contained in the Repair FEC Payload ID of received FEC Repair Packets. A receiver can then compute dw_max_size and derive an appropriate ls_max_size as explained in Appendix C.1.¶

When the observed NSS fluctuates significantly, a FECFRAME receiver may want to adapt its ls_max_size accordingly. In particular when the NSS is significantly reduced, a FECFRAME receiver may want to reduce the ls_max_size too in order to limit computation complexity. A balance must be found between using an ls_max_size "too large" (which increases computation complexity and memory requirements) and the opposite (which reduces recovery performance).¶

C.3. Case of a Non-Real-Time Flow

Finally there are configurations where a source ADU flow has no real-time constraints. FECFRAME and the FEC schemes defined in this document can still be used. The choice of appropriate parameter values can be directed by practical considerations. For instance, it can derive from an estimation of the maximum memory amount that could be dedicated to the linear system at a FECFRAME receiver, or the maximum computation complexity at a FECFRAME receiver, both of them depending on the ls_max_size parameter. The same considerations also apply to the FECFRAME sender, where the maximum memory amount and computation complexity depend on the ew_max_size parameter.¶

Here also, the NSS value contained in FEC Repair Packets is used by a FECFRAME receiver to determine the current coding window size and ew_max_size by observing its maximum value over the time.¶