Methods of queueing theory are often used for telecommunication systems performance evaluation. According to this approach, all MAC and PHY layer protocols details are modelled with a given service time distribution, and the precision of the estimated properties depends on this distribution function selection. While for Ethernet and wireless relay networks the service time is roughly equal to the sum of packet and headers sizes divided by the channel bitrate, it is not so easy to estimate the service time for wireless networks channels based on CSMA/CA including IEEE 802.11 (WiFi) and IEEE 802.15.4 (ZigBee) due to random exponential backoff and collisions. In this paper we use a simple model of the saturated channel described as a semi-Markov random absorbing process, which is based on Bianchi model [2]. We use this process to find a service time phase-type (PH) distribution matching the first two moments of the semi-Markov process. We also use channel simulation model to sample transmission delays in unsaturated mode and find another PH-distribution using G-FIT [12] algorithm. Then we use these PH-distributions to estimate end-to-end delays, queue sizes and nodes utilization in a multi-hop wireless network with linear topology using M/PH/1/N→∙/PH/1/N∙⋯∙/PH/1/N queueing network. All results are compared to wireless network simulation model, and demonstrate the cases in which the queueing network tends to provide results close to real network performance.