A Digital Audio Broadcasting System

T h e audio quality, robustness and complexity issues of a novel mobile digital audio broadcast (DAB) scheme a re addressed. T h e audio codec is based on a combination of subband coding (SBC) and mult ipulse excited linear predictive coding ( M P L P C ) , where the bi t allocation is dynamically adapted according t o both the signal power in different subbands and a perceptual hearing model. Typically a segmental signal t o noise ratio (SEGSNR) in excess of 30 d B associated with high fidelity (HI-FI) subjective qual i ty was achieved for 2.67 bits/sample transmissions a t a mono bi t ra te of 86 kbits/s. Four different source-matched forward error correction (FEC) schemes were investigated in order to explore the complexity, bit rate and robustness trade-offs. W h e n using 4 bit /symbol 16-level star-constellation quadra tu re ampl i tude modulation (16-StQAM) the overall signalling ra te became approximately 30 kBd, accommodat ing two stereo D A B channels in a conventional 200 kHz analogue F M channel's bandwidth. O u r diversity assisted D A B scheme required a channel signal t o noise ratio (SNR) of about 25 d B for unimpaired audio quali ty via the worst-case Rayleigh fading mobile channel, when the mobile speed was 30 m p h a n d the propagation frequency was 1.5 GHz. I n case of the stationary Gaussian scenario a n S N R of abou t 20 d B was required. 1. DIGITAL A U D I O BROADCASTING Analogue frequency modulated (FM) radio broadcasting is antiquated and there is a growing demand for higher quality digital audio broadcasting (DAB) for mobile receivers [I], [2]. Adavanced features, such as five-channel surround sound with ambient-dependent dynamic control, catering for example for reduced dynamic range in a noisy vehicle or trafficand control-data decoding are also desirable features. In this contribution we propose a DAB scheme for mobile channels, which is based on a subband split modified multipulse excited linear predictive (SB-MMPLPC) codec studied in Section 2. The audio bits are protected by a variety of block codes and transmitted using 4 bits/symbol 16-level quadrature amplitude modulation (16QAM) as disMOMUC-2, 2ND INTERNATIONAL WORKSHOP ON MOBILE MULTIMEDIA COMMUNICATIONS, 11-13 APR., 1995, BRISTOL, UK @IEEE cussed in Section 3, while performance figures are reported in Section 4. 2. THE A U D I O C O D E C T h e M M P L P C codec's schematic diagram is shown in Figure 2, which is similar to that of a conventional MPLPC arrangement 1:3], except that it incorporates Nu number of different excitation modes, where we have opted for Nu = 2, corresponding to Mode 1 and M o d e 2 . The audio input signal s (n) is divided into frames of N samples for LPC analysis and the LPC filter parameters are determined by minimising the mean squared prediction error Ew over this interval. Each frame is further divided into contiguous subframes of N , samples, for which the long term predictor (LTP) delay D and gain GI parameters minimising the mean squared error Ew for the current subsegment are determined [4]. The MMPLPC codec's efficiency can be further improved, if the human ear's frequency and energy sensitive p rop erties are exploited by dividing the audio bandwidth into subbands corresponding to the critical bands found in hearing. However, after band splitting, the correlation between adjacent time domain samples is reduced, and the more the band is split, the more this correlation is decreased. The MMPLPC codec utilizes linear prediction requiring high correlation between adjacent samples. In order to compromise, we chose four-band splitting. I n the S B M M P L P C codec seen in Figure 1 the input audio signal s,(n) is split into four subbands: 0-4 kHz, 4-8 kHz, 8-12 kHz, 12-16 kHz, by a Quadrature Mirror Filter (QMF) bank, using two cascaded 64th order QMF filters [3]. The four subband signals {sk(n) , k = 1,2 ,3 ,4) are each encoded by an MMPLPC codec. Since hearing sensitivity is different for the different subbands, the short time energy a: in each subband was estimated and subband k was assigned to one of sixteen empirically designed different bit allocation classes C,, j = 1 . . .16, as demonstrated by Table 1 designed for subbands SB1 and SB2. Similar tables were constructed for the less significant subbands SB3 and SB4 summarising for both excitation modes the number of excitation pulses N , , their quantisation accuracies in terms of the number of bits/pulse as well as the number of bits needed for the encoding of their positions, when using the enumerative method [5]. For the same bit allocation class CJ (k) the lower frequency subbands SB1 and SB2 k = 1,2 were typically assigned a higher number of excitation pulses and higher number of pulse amplitude