160 likes | 179 Views
Explore methods to optimize packet loss repair in VoIP for improved voice quality under bursty loss conditions using FEC and LBR techniques. Comparison with E-Model and implications for future work.
E N D
Comparison and Optimization of Packet Loss Repair Methods on VoIP Perceived Quality under Bursty Loss Wenyu Jiang , Henning Schulzrinne 2002.11.12 이주경
Abstract • VoIP • Packet loss degrades the perceived quality of voice of IP • Packet loss tends to come in bursts • Gilbert loss model • Packet interval • FEC, LBR • Present a method of optimizing the packet interval
Introduction • Packet Loss Repair and Recovery • FEC • LBR(Lower Bit-rate Redundancy) • Redundant data but lower quality version of the same audio • MOS(Mean Opinion Score) • Common VoIP quality metric • 1 ~ 5 :bad, poor, fair, good, excellent
LOSS MODELING • The Gilbert Model
LOSS MODELING(con’t) • Loss Burstiness vs. FEC Performance
Related work : THE E-MODEL • E-model • Analytical model for predicting voice quality • Impairment factor • Delay, loss, echo, loudness, frequency • Each factor is mapped to a score
MOS TEST EXPERIMENT DESIGN • Object • Random vs. bursty(Gilbert) loss model • Compare FEC and LBR, mostly under Gilbert loss • MOS with or without FEC under a wide range of loss probabilities(pu), loss burstiness(pc) and packet intervals(T)
MOS TEST EXPERIMENT DESIGN • Design of an Optimal LBR mechanism • LBR • Main audio codec decoder state drift • Packet alignment order - optimized LBR • Main audio codec packet loss시 • Redundant audio decoding • Reencoding it using a duplicate main encoder • Finally decoding it again using the main decoder • Packet alignment order
MOS TEST RESULTS • Test Set N1: Random vs. Bursty Loss FEC(R) FEC(R) LBR(R) FEC(B) FEC(B) LBR(R) LBR(B) LBR(B) LBR(optimal) LBR
MOS TEST RESULTS(con’t) • Test N1 : Quality of FEC vs. LBR AMR+LBR Figure 13 - MOS(FEC) > MOS(LBR) - bit exact form - sudden switch between low and high audio quality Figure 14 - FEC(2,1) code has best quality
MOS TEST RESULTS(con’t) • Test N2 : MOS Quality vs. Loss Burstiness and Packet Interval • Without FEC • Different Packet interval • (a), (b) • Different Burstness • (c), (d)
MOS TEST RESULTS(con’t) • Test N2 : MOS Quality vs. Loss Burstiness and Packet Interval • MOS of FEC vs. Packet Interval T
MOS TEST RESULTS(con’t) • Comparison with the E-model MOS R value R= 92.4- Id - Ie
MOS TEST RESULTS(con’t) • Optimizing Packet Interval with Delay Impairment
Conclusion and Future work • Evaluation study on the effect of random and bursty packet loss • Generally : MOS(random loss) > MOS(bursty loss) • MOS(FEC) vs. MOS(LBR) • Larger packet interval improves FEC quality • Based on E-model • Trade off between FEC delay and listening quality • Future Work • Determining the reason of MOS test results • Accurate FEC MOS test • FEC : bandwidth overhead and delay