In this report we use Network Utility Maximization (NUM) Framework to study the joint assignment of rate and reliability among sources in multi-hop wireless network. Formulating the above problem in the NUM framework can lead to natural functional decomposition into layers and distributed algorithms. The optimality is preserved through messages passing between layers and local messages exchanging among nodes. The framework reasons (joint) cross layer design. Heuristic layering turns out to be solving a NUM optimization problem. This is so called “Layering as optimization decomposition”. Especially, our problem leads to congestion control at transport layer and the adaptive coding at the physical layer. The messages passing between layers are congestion prices (Lagrange multipliers). Updating is performed distributively using local messages. Each source s has a utility function ππ (π₯π , π π ), where π₯π is an information data rate and π π is the reliability of source s. We assume that the utility function is a continuous, increasing, and concave function of π₯π and π π . Each source s has its information data rate bounded between a minimum and a maximum: π₯π πππ and π₯π πππ₯ , and has a minimum reliability requirement π π min .The reliability of source s is defined as π π = 1 − π π π‘π,π = π₯π /ππ,π ππ,π = πΈπ (ππ,π ) π π = 1 − ∏ (1 − ππ,π ) = 1 − ∏ (1 − πΈπ (ππ,π )) π∈πΏ(π ) π∈πΏ(π ) Assume ππ,π βͺ 1 π π ≈ ∑ ππ,π = ∑ πΈπ (ππ,π ) π∈πΏ(π ) π∈πΏ(π ) π π ≈ 1 − ∑ πΈπ (ππ,π ) π∈πΏ(π ) π π is the end-to-end error probability ππ,π is the code rate of source s at link l π₯π is the information data rate of source s π‘π,π is the transmission data rate of source s at link l ππ,π is the error probability of source s at link l πΈπ (ππ,π ) is an increasing function of ππ,π reflecting the rate-reliability trade-off πΏ(π ) is the set of links used by source s ∑ π‘π,π = ∑ π ∈π(π) π ∈π(π) π₯π ≤ πΆππππ₯ , ∀π ππ,π π(π) is the set of sources using link l πΆππππ₯ is the capacity of link l Assume ππ,π = ππ , ∀π ∈ π(π), ∀π maximize ∑π ππ (π₯π , π π ) subject to π π ≤ 1 − ∑πππΏ(π ) πΈπ (ππ ), ∀π π₯π ∑ ≤ πΆππππ₯ , ∀π ππ π ∈π(π) π₯π πππ ≤ π₯π ≤ π₯π πππ₯ , ∀π π π πππ ≤ π π ≤ 1, ∀π 0 ≤ ππ ≤ 1, ∀π πΏ(π±, π, π«, π, π) = ∑ ππ (π₯π , π π ) + ∑ ππ (ππ πΆππππ₯ − ∑ π₯π ) π π π ∈π(π) + ∑ ππ (1 − ∑ πΈπ (ππ ) − π π ) π πππΏ(π ) = ∑ (ππ (π₯π , π π ) − ∑ ππ π₯π − ππ π π ) π πππΏ(π ) + ∑ (ππ ππ πΆππππ₯ − ∑ ππ πΈπ (ππ )) + ∑ ππ π π ∈π(π) π = ∑(ππ (π₯π , π π ) − ππ π₯π − ππ π π ) + ∑(ππ ππ πΆππππ₯ − π π πΈπ (ππ )) π π + ∑ ππ π ππ = ∑πππΏ(π ) ππ and π π = ∑π ∈π(π) ππ ππ : the Lagrange multiplier on link l with an interpretation of “congestion price”, the price per unit rate to use link l ππ : the Lagrange multiplier on source s with an interpretation of “reliability price”, the price per unit reliability that the source s must pay to the network ππ : with an interpretation of “end-to-end congestion price” on source s π π : with an interpretation of “aggregate reliability price” paid by sources using link l The Lagrange dual function is π(π, π) = max π± min βΌ π± βΌ π± max πmin βΌ π βΌ π πβΌπ«βΌπ πΏ(π±, π, π«, π, π) The dual problem is minimize π(π, π) subject to π β½ π πβ½π Since the Lagrangian is separable, this maximization of the Lagrangian over (π±, π, π«) can be conducted in parallel at each source s maximize ππ (π₯π , π π ) − ππ π₯π − ππ π π subject to π₯π πππ ≤ π₯π ≤ π₯π πππ₯ π π πππ ≤ π π ≤ 1 and on each link l maximize ππ ππ πΆππππ₯ − π π πΈπ (ππ ) subject to 0 ≤ ππ ≤ 1 dual problem can be solved by using the gradient projection algorithm as + ππ (π‘ + 1) = [ππ (π‘) − π½(π‘)(ππ (π‘)πΆππππ₯ − ∑ π₯π (π‘))] , ∀π π ∈π(π) + ππ (π‘ + 1) = [ππ (π‘) − π½(π‘)(1 − ∑ πΈπ (ππ (π‘)) − π π (π‘))] , ∀π πππΏ(π ) π½(π‘) is the step size. Algorithm 1 for Integrated Dynamic Reliability Policy Source problem and reliability price update at source s: ο¬ Source problem maximize ππ (π₯π , π π ) − ππ (π‘)π₯π − ππ (π‘)π π subject to π₯π πππ ≤ π₯π ≤ π₯π πππ₯ π π πππ ≤ π π ≤ 1 ο¬ Price update ππ (π‘ + 1) = [ππ (π‘) − π½(π‘)(π π (π‘) − π π (π‘))]+ where π π (π‘) = 1 − ∑πππΏ(π ) πΈπ (ππ (π‘)) is the end-to-end reliability at iteration t. Link problem and congestion price update at link l: ο¬ Link problem maximize ππ (π‘)ππ πΆππππ₯ − π π (π‘)πΈπ (ππ ) subject to 0 ≤ ππ ≤ 1 ο¬ Price update ππ (π‘ + 1) = [ππ (π‘) − π½(π‘)(ππ (π‘)πΆππππ₯ − π₯ π (π‘))]+ where π₯ π (π‘) = ∑π ∈π(π) π₯π (π‘) is the aggregate information rate on link l at iteration t. In Algorithm 1, to solve problem (8), source s needs to know ππ (π‘), the sum of ππ (π‘)’s of links that are along its path πΏ(π ).This can be obtained by notification from the links, e.g., through the presence or timing of acknowledgment packets in TCP. To carry out price update (9), the source needs to know the sum of error probabilities of the links that are along its path [i.e., its own reliability that are offered by the network π π (π‘)].This can be obtained by the notification from the destination that can measure its end-to-end reliability. To solve link problem (10), each link needs to know π π (π‘), the sum of ππ (π‘)’s from sources π ∈ π(π) using this link. This can be obtained by the notification from these sources. To carry out price update (11), the link needs to know π₯ π (π‘) , the aggregate information data rate of the sources that are using it. This can be measured by the link itself. a linear topology consisting of four links and eight users min(1−πΌ) ππ (π₯π , π π ) = ππ π₯π 1−πΌ − π₯π max(1−πΌ) π₯π πΌ = 1.1, π₯ππππ = 0.1 ( Mb s and π ππππ = 0.9 1 ππ = exp(−π(1 − ππ )) 2 min(1−πΌ) + (1 − ππ ) min(1−πΌ) − π₯π Mb ) , π₯ππππ₯ = 2 ( s π π 1−πΌ − π π max(1−πΌ) π π Mb ) , πΆππππ₯ = 2 ( s min(1−πΌ) − π π ) , π ππππ₯ = 1, maximize ∑π ππ (π₯π , π π ) subject to π π ≤ 1 − ∑πππΏ(π ) πΈπ (ππ,π ), ∀π π₯π ∑ ≤ πΆππππ₯ , ∀π ππ,π π ∈π(π) π₯π πππ ≤ π₯π ≤ π₯π πππ₯ , ∀π π π πππ ≤ π π ≤ 1, ∀π 0 ≤ ππ,π ≤ 1, ∀π, π ∈ π(π) By introducing the auxiliary variables ππ,π , which can be interpreted as the allocated transmission capacity to source s at link l, maximize ∑π ππ (π₯π , π π ) subject to π π ≤ 1 − ∑πππΏ(π ) πΈπ (ππ,π ), ∀π π₯π ≤ ππ,π , ∀π, π ∈ π(π) ππ,π ∑ ππ,π ≤ πΆππππ₯ , ∀π π ∈π(π) π₯π πππ ≤ π₯π ≤ π₯π πππ₯ , ∀π π π πππ ≤ π π ≤ 1, ∀π 0 ≤ ππ,π ≤ 1, ∀π, π ∈ π(π) 0 ≤ ππ,π ≤ πΆππππ₯ , ∀π, π ∈ π(π) We introduce another layer, i.e., link layer (ππ,π ), by using a vertical decomposition of the optimization problem. ′ It is a GP, let π₯π ′ = log π₯π , ππ (π π₯π , π π ) = ππ ′ (π₯π ′ , π π ) maximize ∑π ππ ′ (π₯π ′ , π π ) subject to π π ≤ 1 − ∑πππΏ(π ) πΈπ (ππ,π ), ∀π π₯π ′ − log ππ,π ≤ log ππ,π , ∀π, π ∈ π(π) ∑ ππ,π ≤ πΆππππ₯ , ∀π π ∈π(π) π₯′πππ ≤ π₯′π ≤ π₯′πππ₯ , ∀π π π πππ π π ≤ π π ≤ 1, ∀π 0 ≤ ππ,π ≤ 1, ∀π, π ∈ π(π) 0 ≤ ππ,π ≤ πΆππππ₯ , ∀π, π ∈ π(π) πΏ(π± ′ , π, π«, π, π, π) = ∑ π′π (π₯′π , π π ) + ∑ ππ (1 − ∑ πΈπ (ππ,π ) − π π ) π π πππΏ(π ) + ∑ ∑ ππ,π (log ππ,π + log ππ,π − π₯π ′ ) π π ∈π(π) = ∑ (π′π (π₯′π , π π ) − ∑ ππ,π π₯′π − ππ π π ) π πππΏ(π ) + ∑ ( ∑ (ππ,π ( log ππ,π + log ππ,π ) − ππ πΈπ (ππ,π ))) + ∑ ππ π π ∈π(π) π The Lagrange dual function is π(π, π) = π±′min max βΌ π±′ βΌ π±′max πΏ(π± ′ , π, π«, π, π, π) πmin βΌ π βΌ π πβΌπ«βΌπ π∈C πΆ = {ππ,π | ∑ ππ,π ≤ πΆππππ₯ , ∀π, 0 ≤ ππ,π ≤ πΆππππ₯ , ∀π, π ∈ π(π)} π ∈π(π) The dual problem is minimize π(π, π) subject to π β½ π πβ½π Algorithm 2 for Differentiated Dynamic Reliability Policy Source problem and reliability price update at source s: ο¬ Source problem maximize ππ (π₯′π , π π ) − ππ (π‘)π₯′π − ππ (π‘)π π subject to π₯π ′πππ ≤ π₯′π ≤ π₯′πππ₯ π π π πππ ≤ π π ≤ 1 ο¬ Price update ππ (π‘ + 1) = [ππ (π‘) − π½(π‘)(π π (π‘) − π π (π‘))]+ where π π (π‘) = 1 − ∑πππΏ(π ) πΈπ (ππ,π (π‘)) is the end-to-end reliability at iteration t. Link problem and congestion price update at link l: ο¬ Link problems Link-layer problem maximize ∑π ∈π(π) ππ,π (π‘) log ππ,π subject to ∑π ∈π(π) ππ,π ≤ πΆππππ₯ ; 0 ≤ ππ,π ≤ πΆππππ₯ , π ∈ π(π) Physical-layer problem for source s, π ∈ π(π) maximize ππ,π (π‘)log πππ − ππ (π‘)πΈπ (ππ,π ) subject to 0 ≤ ππ,π ≤ 1 ο¬ Price update + ππ,π (π‘ + 1) = [ππ,π (π‘) − π½(π‘)(log ππ,π (π‘) + log ππ,π (π‘) − π₯π ′ (π‘))] , π ∈ π(π) Reference: [1] Lee, J.-W.; Mung Chiang; Calderbank, A.R., "Price-based distributed algorithms for rate-reliability tradeoff in network utility maximization," Selected Areas in Communications, IEEE Journal on , vol.24, no.5, pp. 962-976, May 2006 [2] M. Chiang “Balancing transport and physical layer in wireless multihop networks: Jointly optimal congestion control and power control,” IEEE J. Sel. Areas Commun., vol. 23, pp. 104, Jan. 2005. [3] Mung Chiang; Low, S.H.; Calderbank, A.R.; Doyle, J.C., "Layering as Optimization Decomposition: A Mathematical Theory of Network Architectures," Proceedings of the IEEE , vol.95, no.1, pp.255-312, Jan. 2007 [4] Palomar, D.P.; Mung Chiang, "A tutorial on decomposition methods for network utility maximization," Selected Areas in Communications, IEEE Journal on , vol.24, no.8, pp.1439-1451, Aug. 2006 [5] Lee Jang-Won; Tang Ao; Huang Jianwei; Mung Chiang; Robert, A., "Reverse-Engineering MAC: A Non-Cooperative Game Model," Selected Areas in Communications, IEEE Journal on , vol.25, no.6, pp.1135-1147, August 2007 [6] Jang-Won Lee; Mung Chiang; Calderbank, A.R., "Utility-Optimal Random-Access Control," Wireless Communications, IEEE Transactions on , vol.6, no.7, pp.2741-2751, July 2007 [7] Jang-Won Lee; Chiang, M.; Calderbank, R.A., "Jointly optimal congestion and contention control based on network utility maximization," Communications Letters, IEEE , vol.10, no.3, pp. 216-218, Mar 2006