IEICE TRANS. COMMUN., VOL.E89–B, NO.7 JULY 2006 2071 LETTER A Study on Power and Bit Assignment of Embedded Multi-Carrier Modulation Schemes for Hierarchical Image Transmission over Digital Subscriber Line Charl` ene GOUDEMAND † , Nonmember, Franc ¸ois-Xavier COUDOUX †a) , Member, and Marc GAZALET † , Nonmember SUMMARY In this letter, we study the problem of designing an effi- cient power and bit allocation scheme in the context of a hierarchical im- age transmission system based on an embedded multi-carrier modulation (EMCM) scheme over digital subscriber line. Authors describe a novel algorithm that performs power minimization under bit rate constraint and QoS requirement. It is based on the Hughes-Hartogs algorithm, and suc- cessively allocates the bits of the high, then low priority data streams. Sim- ulations that assess the performance of the proposed algorithm are also pro- vided and discussed; they demonstrate the interest of the proposed scheme. key words: hierarchical modulation, digital subscriber line, embedded multi-carrier modulation, image transmission 1. Introduction Recently, hierarchical transmission has been investigated in order to improve the performance of digital image and video communications [1], [2]. The idea of hierarchical transmis- sion is to transmit information with varying reliability ac- cording to its importance. A hierarchical transmission sys- tem is composed of a hierarchical source coder and corre- sponding channel coder; it divides the information into sev- eral layers (here, two layers) according to their significance, and transmits each layer with different reliability according to the layers. Embedded Multi-carrier Modulation (EMCM) scheme using hierarchical Quadrature Amplitude Modula- tions (QAM) can be used as a hierarchical channel coder. In this case, a power and bit loading algorithm allows optimiz- ing the transmission bandwith as well as the power distribu- tion among the different sub-channels [3]. In [4], Pradhan et al. have proposed a bit and power as- signment method to maximize a weighted sum R w of base layer (R 1 ) and refinement layer (R 2 ) rates: R w = μR 1 + (1 - μ)R 2 , subject to a power constraint and bit error rate (BER) requirements. By varying μ we get a trade-off between the base layer and refinement layer throughput rates. In prac- tice, however, the μ parameter value is unknown, and it is difficult to adjust the throughput rates in order to match the optimized value, because both rates are fixed a priori by the encoding parameters of the hierarchical source coder. Hence, the loading algorithm in [4] is not well suited to prac- tical image transmission schemes. Manuscript received December 19, 2005. † The authors are with IEMN-DOAE (UMR CNRS 8520), Uni- versity of Valenciennes, France. a) E-mail: coudoux@univ-valenciennes.fr DOI: 10.1093/ietcom/e89–b.7.2071 In this letter, we study a new bit and power assignment method that is more realistic and better suited to the EMCM image transmission framework. The algorithm minimizes the total power consumption subject to a bit rate constraints and BER requirements for the two layers. It is based on the Hughes-Hartogs allocation method [5]; the principle is to al- locate the bits successively to each of the high-priority and low-priority layers. Finally, simulation results of hierarchi- cal video transmission over digital subscriber line are pro- vided and compared to the ones obtained with the scheme described in [4], as well as a optimal frequency-division- multiplexed (FDM) based scheme described in [6]. 2. Proposed Algorithm for Power and Bit Allocation In order to realize a two layers EMCM scheme we use a hierarchical source coder which splits the video source into base information and refinement information. We suppose that the base and refinement information bit rates are R 1 and R 2 , respectively. These two layers are associated to dif- ferent informational content with varying levels of signifi- cance. For example, the base information contains motion vectors and DC coefficients that have to be transmitted reli- ably to the receiver. The refinement information consists in high-frequency DCT coefficients that have limited impact on visual quality. Hence, we set unequal levels of channel error protection of the base and refinement layers, B 1 and B 2 , respectively. The base and refinement information are fed into the hierarchical channel coding based on the EMCM scheme using hierarchical modulation. In the modulator, hierarchi- cal QAMs are assigned to each sub channel, so that the same sub channel carries simultaneously the bits of both base and refinement layers. Figure 1 shows some examples of hierar- chical QAM constellations. Here, base information bits are assigned to the clusters, and the refinement information bits are assigned to the satellites. The distances d 1 and d 2 are un- equal and adjusted to meet the desired BERs. The higher the distance d 1 , the more protected the base layer; but the less protected the refinement layer, due to the limited power con- straint. In what follows, we consider the use of hierarchical N/ M QAMs, with N = 2 n , n being equal to the number of base information bits, and M = 2 m+n , m being equal to the number of refinement information bits. Practically, the pa- Copyright c  2006 The Institute of Electronics, Information and Communication Engineers