Adaptive LeNet-assisted multi-objective energy optimization for massive MIMO-integrated NOMA edge computing systems

One of the main challenges in improving efficient energy in Mobile Edge Computing (MEC) lies in the time needed to generate training data. Conventional approaches rely on computationally intensive algorithms to determine optimal power allocation in consuming time and large-scale networks are become impractical with thousands of channel conditions. Efficient data generation is therefore essential to support fast and reliable predictions that enables energy-efficient task offloading in dynamic MEC environments. So, this work proposes an Energy-efficient Resource Allocation framework for MEC networks integrating Non-Orthogonal Multiple Access (NOMA) and massive Multiple-Input Multiple-Output (MIMO) (ERA-MNM) framework. Multiple user devices consist a base station (BS) and a cloudlet with limited computational capacity. Each user has the option in computational tasks to derive locally or offload them to a cloudlet through the BS. The NOMA and MIMO frameworks together optimize power allocation, subchannel assignment, task offloading, and user grouping to improve the overall system performance. Users are assigned to subchannels based on the highest channel gains and paired user groups shared its same subchannel using NOMA principles, while MIMO technology enables spatial multiplexing to support efficient data transmission. A customized LeNet (Cus-LN)-based neural network is employed to intelligently perform task offloading decisions and predict optimal power allocation for both MIMO and NOMA scenarios. The network processes input features with convolutional layers, activation functions and pooling operations that extracts hierarchical features. By jointly considering local computation, task offloading and NOMA-MIMO power allocation, the system enhances spectral efficiency, reduces energy consumption and improves overall computation offloading performance. The Cus-LN has delivered lesser energy by 190 mJoules for 40 users at the BS and 400 mJoules for 60 users at the BS over the delay condition of 400 ms.