R. SuganyaL. R. SujithraDr. Ramesh Kumar AyyasamyP. Chinnasamy2025-10-312025-10-312025-0210.1002/ett.70071https://dspace-cris.utar.edu.my/handle/123456789/11682In fifth-generation (5G) radio access networks (RANs), network slicing makes it possible to serve large amounts of network traffic while meeting a variety of demanding quality of service (QoS) standards. Higher path loss and sparser multipath components (MPCs) are the primary distinctions, which lead to more notable time-varying characteristics in mmWave channels. Using statistical models, such as slope-intercept methods for path loss for delay spread and angular spread, is challenging to characterize the time-varying properties of mmWave channels. Therefore, adopting mmWave communication systems requires highly accurate channel modeling and prediction. This research proposes a novel technique in wireless mmWave communication 5G network slicing and routing protocol using IoT (Internet of things) and deep learning techniques. An adaptive software-defined reinforcement recurrent autoencoder model (ASDRRAE) slices the mmWave communication network. A dilated clustering-based adversarial backpropagation model (DCAB) then performs network routing. The experimental analysis evaluates throughput, packet delivery ratio, latency, training accuracy, and precision. The suggested hybrid model has a 97.21% overall recognition rate, illustrating that the suggested strategy is aptly applicable. A 10-fold stratified cross-validation is employed to evaluate the suitability of the proposed method. © 2025 John Wiley & Sons Ltd.en5G network slicingdeep learning techniquesIoTrouting protocolwireless mmWave communicationDeep reinforcement learningInternet of thingsInternet protocolsRadio access networksRouting protocolsSoftware radio5g network slicingDeep learning techniqueLearning techniquesMm wavesMm-wave CommunicationsNetwork slicingPath lossRouting modelRouting-protocolWireless mmwave communication5G mobile communication systemsjournal-article