This work is an establishment for routing enhancement in wired and wireless data communications. The work is classified into three distinct parts. Part I is focused on the IP packet networks. Part II concerns routing and interference handling in cellular communications. Part III illustrates our scheme of collision handling in TDMA-based and ad hoc networks. \textbf{Wired communications:} Relay routing for IP networks has been well-documented in past years. However, the implementation cost of relay solutions has not yet been conclusively identified. Dynamic relay relies on periodic probing of routing changes for enhanced performance. Static relay routing, however, uses nonperiodic probes to measure latency and packet loss at a lower rate. The ultimate goal of a dynamic relay is to maximize QoS, while a static relay seeks reliability in routing over time. There exists considerable research focused on understanding IP link changes due to fluctuations in routing dynamics. This work, in particular, examines new relay characteristics such as the number of hops in a relay path or Hop-To-Live (HTL) of minimum delay relay paths over similar dynamics. This HTL is to predict minimum and stable relay paths and minimize the probing overhead. By considering delay, we find those relay paths are more stable and short. This study proposes a hybrid TCP-UDP for TCP-based services such as YouTube and VoIP. The analysis of streaming of estimation-based layer-$ 3 $ hybrid relay on a network of $ 140 $ shows a bandwidth increase for such streams. Further, we have included an analysis of $ 18,906 $ delay traces from a network of 138 hosts to demonstrate the rich existence of IP relay paths that can be leveraged to enhance layer-$ 3 $ performance. This is a counterpart of current and pure TCP relay schemes and benefits delay and throughput-sensitive applications. The specific analysis of end-to-end delay is to enhance TCP throughput and transfer time over the default TCP relay. Results show such relay benefit TCP by orders of magnitude. \textbf{Cellular communications:} There is a new trend in cellar networks that allows mobile carriers for dynamic spectrum access. The FCC adopted rules for shared commercial use of the $ 3550 $-$ 3700 $ MHz band, Citizen Broadband Radio Service (CBRS) in 2015. This band is occasionally in radar data communications and is now open for future LTE-based dynamic-spectrum access. Further, currently, there is no infrastructure corporation between cellar carriers like Verizon Wireless, AT\&T, T-Mobile, or Sprint for instance, in normal or emergencies. Dynamic spectrum access is proposed to better utilize the scarce spectrum resources. Recently, the FCC opened up the Citizen Broadband Radio Service (CBRS) for wireless service providers to enable dynamic spectrum access for 5G networks. However, the lack of collaboration among cellular providers has been hampering reliable access due to the increasing interference from multiple carriers. This work considers a new network architecture to foster collaboration among providers through Collaborative Multihop and Multi-Channel Cellular Networks (CMCNs). This work proposes a suboptimal solution to manage interference and congestion by utilizing minimum backoff schemes. Moreover, we develop a near-optimal scheduling algorithm to provide Minimum Broadcast Delay Forwarding (MBDF), which approximates a polynomial scheduling paradigm. Then, we present 6-policy based scenarios that summarize the scheduling paradigm. The designed mapping paradigm is to compute a near-optimal solution. This mapping via conflict graphs allows our analysis to establish a new transformation for MBDF, on which dependency between interference components can be easily visualized and solved. The simulation results show that our approach outperforms the existing models in terms of scheduling delay and message redundancy. The confidence in our proposed scheduling has been examined over distinct sets of CBRS channels by comparing MBDF with a minimum scheduling demand. \textbf{Ad Hoc communications:} In ad hoc collision is a major cause of performance degradation. TDMA scheduling is a well-studied subject for ad hoc networks to manage collisions. Reinforcement Learning (RL), in particular, Q-Learning (QL) is a tool to achieve a collision-free TDMA schedule. However, the slow convergence and the complex reward design might be a challenge for a distributed QL-based TDMA over a minimum frame length. This work proposes a QL-based TDMA for distributed systems via control from a slot-based weighted exploration algorithm. This new algorithm is a distributed directional and count-based network weighting and cycle detection for slot-based wireless networks with a reduced search overhead comparing existing schemes like Rocha–Thatte algorithm in~\cite{rocha2015distributed}. The proposed QL-based TDMA applies a slot of length $ 2.5 $ milliseconds in a minimum frame length for any given topology. The reasons behind such a modification when using a minimum required frame are: (1) To achieve smaller communication delay. (2) This minimum frame reduces the scheduling handshake overhead or acknowledgments. The literature is rich in conventional TDMA protocols that operate above minimum frame length and are designed as ACK/NACK-based scheduling schemes. Here, nodes learn successful policies for a collision-free frame based on previous collision statistics. This occurs while having minimum NACK and no ACK flows. Each newly chosen slot is accessible only over the next frame and not the current one. This could raise convergence time. However, for a node with a synchronized and just elapsed current sending slot of a frame and a newly chosen slot of the next frame, such a node can still examine all slots between those two slots of the current frame instead of the next one to minimize convergence. This work has shown how wireless nodes could benefit from collided signals to avoid the next collisions, and while being in a listening mode can infer $ 1 $ hop and $ 2 $ hops surrounding collisions. The simulation has shown that wireless sensor nodes can learn minimum collision-free frames over a practical convergence period.
Read
