Adaptive Rate Control for Streaming Stored Fine- Grained Scalable Video Philippe de Cuetos, Keith W. Ross NOSSDAV 2002, May 12-14,2002

Outline Introduction Proposed Framework Problem Formulation Optimal Transmission policy Real-time rate adaptation algorithm Experimental Result

Introduction Small time-scale bandwidth fluctuations Maintaining a small Play-back Delay Buffers at the client store the video frames before they are decoded Longer time-scale bandwidth fluctuations Using multiple versions of the video or a layered encoded video To maximize the overall quality

Introduction (cont.) The goal in this paper is to develop low- complexity yet high-performing scheme that adequately adapt to the short- and long-term variations in available bandwidth. Using the streaming of 2-layer FGS videos.

Framework - Server rbrb rere K s (t)*r e Xe(t)Xe(t) Xb(t)Xb(t) X(t)X(t) Base Layer Bitstream Enhancement Layer Bitstream

Bandwidth - Server The rate at which the server transmits frames into the network depends on the available bandwidth during the slot We divide the video into several slots, and the slot length is equal to a constant C

Framework - Client Decoder EL Buffer BL Buffer Xe(t)Xe(t) Xb(t)Xb(t) X(t)X(t) Yb(t)Yb(t) Ye(t)Ye(t) rbrb K c (t)*r e

Framework - Feature Client buffers needed: Y b (t), Y e (t) Prefetching throughout playback :∆ 0 Losses may only occur due to missed deadlines X(t): the available bandwidth at t X(t) is known in the beginning Synchronous transmission across base and enhancement layers

Framework - Feature ∆(t) the number of seconds of video contained in the buffer in time t ∆(t) = Y b (t) / r b

Notations

Transmission policy A set of successive video encoding rates The restriction to transmission policy Ensure a minimum of quality, by ensuring the decoding of the BL data without loss Maximize the bandwidth efficiency Minimize the variations of the rendered coding rate between successive video sequences

Efficiency and Variability Function Bandwidth Efficiency Coding Rate Variability

Optimal Transmission Policy This optimization serves two purpose It provide a useful bound on the achievable performance when the bandwidth is not known This theory help us design an adaptation heuristic for the realistic case when the bandwidth is not known The priority of optimization criteria Base-layer loss Bandwidth efficiency Coding rate fluctuation

Condition for No losses Theorem 1. The transmission policy, (r s (0),….,r s (n-1)) yields no loss of data over the whole decoding duration if and only if, for all k=0, …, n-1, r s (k) <=β k (∆ k ) whenever ∆ k < C, where rs(k)rs(k) rc(k)rc(k) = r s (k)

Condition for No losses (cont.) For every time t  [t k +∆ k, t k+1 ], the total amount of data that the client attempts to consume is less than the amount of data that was transmitted

Maximizing Bandwidth Efficiency According the Bandwidth Efficiency Formula it is equivalent to maximizing t end

Minimizing Rate Variability

Minimizing Rate Variability (cont.) t/2t 1M 0.5M 0 BL BitstreamEL Bitstream X ave (t) = 0.75M t/2 For the bitstream of the next time slot

Minimizing Rate Variability (cont.) Using DP to obtain the cost c k (∆, ∆’) c k (∆, ∆’) will be defined as (r-r’) 2, where r is the value of r s (k) r’, the value of r s (k+1) ∆ ∆’ ∆k∆k ∆ k+1 t = k*C t = (k+1)*C c k (∆, ∆’)

Rate Adaptation Algorithm Provide a heuristic real-time policy that will adapt to the variations of X(t)

Experimental Result

Experimental Result (cont.)