Dead−time and feed−forward disturbance compensation for congestion control in data networks

Abstract

Two major features characterize the congestion control problem in modern data networks: (1) propagation delays, which may have an adverse impact on the stability of closed loop congestion control algorithms, and (2) time−varying available bandwidth. To deal efficiently with these issues, it is proposed to design the controller by employing both dead−time and feed−forward disturbance compensation. Dead−time compensation provides an output that quickly matches the command signal, whereas feed−forward compensation rejects the steady−state error, which is due to time−varying available bandwidth, and provides high−link utilization without large buffering. It is also shown that standard proportional−derivative−integral controllers are not effective in data networks due to large propagation delays. Finally, dead−time compensation is compared with respect to a proportional controller to illustrate major advantages of the controller proposed here.