|
Proposed scheme | Algorithm based on | Cell load metric representation | Inputs to cell association algorithm | Cell-edge UE throughput gain | Average UE throughput gain | Advantages | Drawbacks |
|
Gu et al., 2013 [6] | Practical adaptation based on cell-edge UE capacity | Number of pieces of UE based RBUR | Network acquired | Nearly optimal with 70% gain over static at 6 dB | Not provided | Simplicity and immune to feedback delays and errors | No criteria for selecting the cell load threshold |
|
Gu et al., 2014 [7] | Practical adaptation based on predicted overall capacity | CBR traffic based RBUR | Network acquired based on system capacity prediction | Not provided | −9.4% gain compared to the optimal | Can achieve nearly optimal performance in all cell load conditions | The algorithm updating mechanism is based on trial and error |
|
Kudo and Ohtsuki, 2013 [8] | Q-learning based on the amount of outage reduction | Set of ratio of RBs and UE distribution | Distributed: pieces of UE learn their optimal bias values | 61% gain over no learning scheme at 20% PRB | 200% gain compared to optimal at 40% PRB | Pieces of UE learn its bias values from past experience to maximize network throughput | Long convergence time makes it not suitable for real systems. Effect of UE mobility not considered. |
|
Koizumi and Higuchi, 2013 [9] | Simple adaptation based on expected minimum average UE throughput | A combination of MeNB index and resource index that maximizes UE throughput | Decentralized: no need for coordination among MeNBs | 1.3-fold gain compared to the no ICIC case | 1.3-fold compared to the no eICIC case | Faster convergence, further enhanced performance with eICIC, and adapting according to the variation of UE distribution | Large overheads due to feedback. Ping-pong handover problem with multiple pieces of UE |
|
Kikuchi and Otsuka, 2013 [10] | Adaptive control CRE based on SINR | Ratio of the number of pieces of UE connected to PeNBs and MeNBs | Centralized based on the feedback from UE | Near-optimal performance | Slightly above-optimal performance: 3.3 Mbps compared to optimal with 3.2 Mbps | Simple algorithm. It has the ability to solve the trade-off between cell-edge UE throughput and the average UE throughput | Delay due to feedback from UE. Number of pieces of UE cannot accurately estimate the cell’s load condition |
|