6. SIMULATION RESULTS
6.2. Objective Quality Analysis
The rate-distortion curves of these bit allocation schemes are shown in Fig. 7 and 8. Two major evaluative methodologies, ANMR and MNMR, are used for distortion. We can find that the performance of the CTB scheme is similar to that of the JTB scheme. The ANMR performance loss is less than 0.2dB for One-Loop CTB-ANMR and less than 0.1dB for Two-Loop CTB-ANMR (the lowest three curves in Fig. 7). The MNMR performance loss is less than 0.1 dB for both One- and Two-Loop CTB-MNMR (the lowest three curves in Fig. 8). Both of them are much better than the MPEG-4 VM (the top line).
YANG AND HANG CASCADED TRELLIS-BASED OPTIMIZATION FOR AAC
AES 115TH CONVENTION, NEW YORK, NEW YORK, 2003 OCTOBER 10-13
8
The differences of performance between the fast searching algorithms and the original CTB-MNMR scheme are shown in Fig. 9 and 10. In light of the complexity analyses on Gm_Nu and Lm_Nu, and the uniform NB_SF fast algorithms, with NB_SF=12 and 5, are chosen for comparison. There is nearly no performance loss for the Gm_Nu algorithm (ANMR or MNMR Difference ≈ 0). The advantage of the non-uniform algorithms over the uniform algorithms at about the same complexity is clearly shown in Figs. 9 and 10.
Total Bit Rate (kbps)
16 32 48 64 80
Fig. 7. ANMR Rate-Distortion Analysis
Total Bit Rate (kbps)
16 32 48 64 80
Fig. 8. MNMR Rate-Distortion Analysis 6.3. Subjective Quality Analysis
Listening test by human ears is the traditional method to subjectively evaluate the audio quality and is also the most recognized subjective quality test. However, such subjective test is expensive, time consuming, and difficult to reproduce. Informal listening tests on the aforementioned schemes show that it is hard to differentiate between JTB and various CTB schemes.
In addition, a “simulated” subjective test, Objective Difference Grade (ODG), has been conducted. ODG
is a measure of quality designed to be comparable to the Subjective Difference Grade (SDG). It is calculated based on the difference between the quality rating of the reference and test (coded) signals. The ODG has a range of [-4, 0], in which –4 stands for very annoying difference and 0 stands for imperceptible difference between the reference and the test signals [10][11]. The ODG results for various search schemes discussed in this paper are shown in Fig. 11 and the reference signal is the original audio sequence. According to the collected test data (Fig.
11), the difference between JTB and CTB schemes is quite small. The ODG results, which are relative to the CTB-MNMR scheme, for various fast searching algorithms are shown in Fig. 12. Again the performance of the non-uniform NB_SF algorithms is better than that of uniform NB_SF algorithms at about the same computational complexity.
Total Bit Rate (kbps)
16 32 48 64 80
Fig. 9. ANMR Difference Analysis
Total Bit Rate (kbps)
16 32 48 64 80
Fig. 10. MNMR Difference Analysis
YANG AND HANG CASCADED TRELLIS-BASED OPTIMIZATION FOR AAC
AES 115TH CONVENTION, NEW YORK, NEW YORK, 2003 OCTOBER 10-13
9 Total Bit Rate (kbps)
16 32 48 64 80
Fig. 11. ODG for VM-TLS, JTB and CTB
Total Bit Rate (kbps)
16 32 48 64 80
Fig. 12. ODG for Various Fast Searching Algorithms
7. CONCLUSIONS
In this paper, we propose a CTB optimization scheme for the MPEG-4 AAC coder, in which the optimization procedures for finding coding parameters, SF and HCB, are separated in two consecutive steps. Based on the complexity analysis, the proposed CTB scheme is approximately 71 to 142 times faster than the JTB scheme. Moreover, the simulation results show that both the objective and subjective quality of the CTB scheme is close to that of the JTB scheme. In addition, we also propose a lossless fast searching algorithm for trellis-based HCB optimization, which is about 4 times faster.
Furthermore, two non-uniform searching algorithms, Gm_Nu and Lm_Nu, are proposed for trellis-based SF optimization. The simulation results show that the non-uniform searching algorithms can achieve better performance than uniform searching algorithms under the same complexity.
8. ACKNOWLEDGMENT
This work was supported by National Science Council, Taiwan, R.O.C., under Grant NSC-91-2219-E-009-011.
9. REFERENCES
[1] ISO/IEC JTC1/SC29, “Information technology – vary low bitrate audio-visual coding,” ISO/IEC IS-14496 (Part 3, Audio), 1998.
[2] M. Bosi, et al., “ISO/IEC MPEG-2 advanced audio coding,” Journal of Audio Engineering Society, vol. 45, pp. 789-814, October 1997.
[3] A. Aggarwal, et al., “Trellis-based optimization of MPEG-4 advanced audio coding,” Proc. IEEE Workshop on Speech Coding, pp. 142-4 2000.
[4] A. Aggarwal, et al., “Near-optimal selection of encoding parameters for audio coding,” Proc. of ICASSP, vol. 5, pp. 3269-3272, Jun 2001.
[5] T. V. Sreenvias and M. Dietz, “Vector quantization of scale factors in advance audio coder (AAC),” Proc. of ICASSP, vol. 4, pp. 3641-3644, May 1998.
[6] H. Najafzadeh and P. Kabal, “Improving perceptual coding of narrowband audio signals at low rates,” Proc. of ICASSP, vol. 2, pp. 913-916, March 1999.
[7] H. Najafzadeh and P. Kabal, “Perceptual bit allocation for low rate coding of narrowband audio,” Proc. of ICASSP, vol. 2, pp. 893-896, 2000.
[8] “The MPEG audio web page.” http://www.tnt.
unihannover. de/project/mpeg/audio.
[9] European Broadcasting Union, Sound Quality Assessment Material: Recordings for Subjective Tests Brussels, Belgium, Apr. 1988.
[10] Draft ITU-T Recommendation BS.1387:
“Method for objective measurements of perceived audio quality,” July. 2001.
[11] A. Lerchs, “EAQUAL software”, Version 0.1.3 alpha, http://www.mp3-tech.org/
附錄 C
附錄 D
Rate control for real time media based on predictive wireless channel