Analysis of a cross-layer Hybrid-ARQ scheme: application to unequal packet protection Aude Le Duc *‡ , Philippe Ciblat ‡ , and Christophe J. Le Martret † * ESME-SUDRIA, Paris, France † THALES Communications, Paris, France ‡ TELECOM PARISTECH, Paris, France Abstract— In order to improve the reliability of any HARQ technique at the IP level, a cross-layer strategy between the MAC and the IP levels has been recently developed in [1]. This strategy consists in replacing a retransmisssion credit per MAC packet with a retransmission credit per IP packet. In other words, the transmission credit is shared by the MAC packets belonging to the same IP packet. Packet Error Rate, throughput, delay and jitter for this new approach have been extensively analyzed in [1], [2]. In this paper, we remark and theoretically prove that the first MAC packets are better protected than the last ones. This leads to a natural ”unequal packet loss protection” that can be useful in many applications, such as video streaming, compressed image transmission, audio and protocols (TCP/IP). Exploiting this novel manner of protecting packets unequally may not exclude the use of the standard manners, such as different feedforward error correcting codes (FEC), modulation size, or hybrid ARQ scheme per required quality of protection. Finally, our claims are supported by numerical simulations. I. I NTRODUCTION In real systems, in order to protect the data against transmis- sion errors, it is usual to carry out Feedforward Error Coding (FEC) at the PHY layer and to add an Automatic ReQuest (ARQ) mechanism at the MAC layer. To still improve the transmission reliability, a combination of both approaches can be done and leads to the so-called Hybrid ARQ (HARQ) scheme [3]. This HARQ scheme is located at the MAC layer. By taking into account the presence of the IP layer (in the context of TCP-IP protocol), it is possible to optimize the HARQ scheme in the following way [1]: instead of having a transmission credit per MAC packet, it is of interest to share the total transmission credit among all the MAC packets belonging to the same IP packet. This technique enables us to improve the Packet Error Rate (PER) at the IP level significantly. In this paper, focusing on the MAC level, we will see and prove that the MAC packets belonging to the same IP packet are inherently unequally protected if the above- mentioned cross-layer optimization is considered. Before going further, we recall that it is worthy of protecting the data, and thus the packets, differently for some applica- tions. For instance, in the video streaming context, the data associated with the first image (called usually I-Frame where I stands for Intra-coded) have to be more protected than the data associated with the other images (called usually P-Frames where P stands for Prediction) representing the motion in the video sequence [4], [5]. Similar issue has to be addressed in the context of scalable image where the image is decomposed through a discrete-cosine transform or a wavelet transform for which the data associated with the low-frequency image decomposition are more important than the data corresponding to the high-frequency image decomposition [7], [8]. Equivalent issues arise in voice transmission [9]. Lastly, in IP context, the IP header has to be read correctly. As a corollary, for some other applications (such as the voice), some MAC packets (actually the less important ones) belonging to the same IP packet can be lost without affecting significantly the system performance [10], [11] and thus it can be interesting to forward these partially corrupted IP packets to the upper layer [13]. In the literature, one can find various ways to do unequal packet protection: for instance, a more robust coding scheme is applied to the packets that have to be more protected. If HARQ scheme is carried out, the packet to be more protected is protected with a more powerful HARQ scheme in which the transmission credit may be higher or the underlying coding scheme may be stronger [14]. As already explained before, we exhibit an additional (and so complementary) way for protecting the packets in an unequal manner by using the cross-layer HARQ scheme introduced in [1]. Therefore, we consider that it is of interest to inspect the following problems: • what is the error probability of each MAC packet be- longing to the same IP packet in the framework of [1]? Once the derivations have been done, is each MAC packet really unequally protected? • what is the probability to get a certain number of erro- neous MAC packets out of all the MAC packets belonging to the same IP packet? This paper is organized as follows : in Section II, we introduce the system model and we remind some basic results about derivations of packet error rate. In Section III, we derive the probability to erroneously receive each MAC packet belonging to the same IP packet for the XO strategy. We prove that this probability depends on the location of the considered MAC packets in the IP packet which lead to an unequal packet protection. In Section IV, we derive the probability to successfully receive a given number of MAC packets among the whole set of MAC packets belonging to the same IP packet. Section V is devoted to numerical illustrations. Conclusions are drawn in Section VI.