First, let us define some important terms

1) Transport block (TB) is a group of resource blocks with a common modulation/coding. The physical interface is a transport block, which corresponds to the data carried in a period of time of the allocation for the particular UE.

2) Service data unit (SDU) is the actual data packets that are coming from the top of the protocol stack.

3) Protocol data unit (PDU) is information delivered as a unit among peer entities of networks containing control information, address information or data. In layered systems, PDU represents a unit of data specified in the protocol of a given layer, which consists of protocol control information and user data. several RLC SDUs often packed into a single PDU.

The HARQ process is done in combination between the MAC and the PHY to retransmits transport blocks (TBs) for physical interface error recovery. The PHY performs the retention and re-combination (incremental redundancy or transport block CRC failure check) and the MAC performs the management and signaling by indicating a NACK message when there’s a transport block CRC failure detected by PHY layer. (Note: In HARQ operation, the retransmission does not have to be fully correct. It has to be correct enough that it can be combined mathematically with the previous transport block in order to produce a good transport block.)

ARQ applies to an RLC SDU. As we know, RLC performs segmentation and reassembly of RLC PDU into RLC SDU in three type of operation in modes: transparent mode (TM), acknowledged mode (AM) and unacknowledged mode (UM).

Transparent mode is used only for control plane signaling for a few RLC messages during the initial connection. There is effectively no header; it simply passes the message through. Unacknowledged mode (e.g. voice over IP) and acknowledged modes (e.g. TCP/IP or critical information) use the RLC header which indicate whether or not the ARQ mechanism is involved. So applying ARQ is data or application dependent. Also ARQ and HARQ are complementing each other’s, in a way that when HARQ transmitter detects a failed delivery of a transport block the relevant transmitting ARQ entities are notified and potential retransmissions and re-segmentation are initiated at the RLC layer for the affected PDUs.

A short answer: you can, but that is not the motivation for separate layers. A long answer: H-ARQ was put at a lower layer to make SDU processing time faster, therefore reducing latency (imagine having to wait for the next frame to send a piggybacked ACK in the RLC-AM, or queue up in the buffer because MAC priority queues are busy). Unlike MAC, RLC-AM will discard any frame that is received in error. MAC uses subsequent retransmissions, which happen at a faster rate, to reconstruct the frame. Now I am sure you knew that already, and you are asking why not having H-ARQ and ARQ both done at a new Layer 2 protocol. The answer is: you can, but you will still need a separate layer that performs the encryption, dynamic scheduling, polling, sending status updates, and segmentation. Also, you need to be able to multiplex several logical channels into a single transport channel. All of that without impacting the service time of PDUs awaiting service. If all of the RLC functionalities were implemented in that new Layer 2 protocol, it would cause extra processing complexity which translates to extra latencies. In fact, layer 2 is not only RLC and MAC, but also PDCP is a Layer 2 protocol. Imagine if we were to combine all those layers with their functionalities to add header compression and security to generate one expensive Layer 2 super layer! Therefore, a division of Layer 2 into specialized sublayers is dictated by such a need. I hope this answers your question!

Faris B. Mismar , Buffering of SDU's is also done at PDCP and the packets are only removed once ACK is received from RLC.

I tried to find the logic for this and think that this is mainly done due to following 2 points.

1. be better suitable for the Split architecture for 5G, in case HO between gNB CU's have to be done. If it wasn't buffered at PDCP RLC would have to send the packets back to PDCP.

2. Since PDCP is an anchor point for Dual connectivity and Packet Duplication

Would you happen to know any other reason? please let me know what you think