This section shows the EE performance of the proposed bio-inspired algorithm (, designated as "BIBPA"), compared to the distributed ADMM-based energy-efficient resource allocation algorithm (denoted as "PA"). Consider multiple cells. The radius of each cell is fixed at 500m. In a large MIMO network, the system bandwidth is normalized with B = 1 Hz and the bandwidth of each subcarrier is BS = B / S Hz. IOT devices are randomly and evenly distributed within each cell. Both path loss and Rayleigh fading are taken into account. The simulation results are based on an average of 10,000 iterations. According to [35] some system parameters are listed in Table I.

**Table I Simulation parameters** [35]

Parameter | Value | Parameter | Value |

K | 6 | £ | 0.8 |

U | 15 | λ | 10− 7 |

S | 20 ~ 40 | B | 1 Hz |

Pbs Max | 46dBm | PDAC | 10 mW |

Puser Max | 23dBm | PADC | 10 mW |

α, k,u | 1/75 | Pfilt | 2.5 mW |

Pmin | 30.3 mW | Psyn | 50 mW |

PLNA | 20mW | PIFA | 3 mW |

Figure 1 compares the EE of BIBPA with the EE of PA. The "BIBPA" EE at full CSI turns out to be superior to all other cases, thanks to its highly efficient AS scheme and full channel conditions. Considering the estimation error in the actual channel environment, the EE of "BIBPA" at incomplete CSI is lower than the EE at full CSI and EE of "PA".

Figure 2 shows the difference in performance between a full CSI and an incomplete CSI. At a transmission distance of 100 m, the system EE at full CSI is about 3.5 times the system EE at full CSI with σ2ek, u, s = 0.7.

, if you increase the transmission distance to 200m, the EE at full CSI will be about 1.5 times that of σ2ek, u, s = 0.7. The larger the variance of the channel estimate, the larger the gap between the variances of the difference. For example, the EE of σ2ek, u, s = 0.3 is higher than in both cases, and all observations in Fig. 2 show that the EE of the system depends on the accuracy of the CSI obtained.

Figure 3 shows the number of complete and incomplete CSI EEs and antennas. It can be seen that EE increases first and then decreases with this increase in the number of antennas. For example, with an incomplete CSI (σ2ek, u, s = 0.6), the EE curve increases from about 0.6x1023 bits / j / Hz at Mk = 10. After peaking 2.2 at Mk = 30 at 2.2x103 bits / j /, the "PA" system EE shows a downward trend. The reason is that the system-EE is an appropriate number of antennas that can be effectively improved, but there are too many antennas. Immediately power consumption will increase and EE will decrease. In addition, AS can reduce the hardware accomplices of multi-antenna systems. Therefore, it is necessary to properly determine the optimum number of antennas.

Figure 4 shows EEs with different antenna numbers and CSI estimation errors. As shown in Fig. 4, to achieve maximum EE, there is optimal through power for a particular antenna number and CSI. As the transmit power increases, the system EE increases slightly, then reaches maximum, and finally moderately decreases, wondering about complete or incomplete CSI. This shows that the EE of the system can be improved by properly allocating the transmit power. It also shows the various antenna numbers Nk, u, s = 20, 25, and 40. We also compare that EE is superior to incomplete CSI under complete and incomplete CSI.

To demonstrate the convergence performance of PA, Fig. 5 compares it with the Dinkelbach-based centralized algorithm [36] and the OMA-based scheme using the Lagrange multiplier [37]. Benchmark algorithms are called Dinkelbach-based CA [38] and "LAOMA [36]". Figure 5 shows the convergence of PA and the effect of the convergence parameter p. This shows that the optimal EE can be reached with a small number of iterations by adjusting ADMM when p = 0.068, and the optimal EE can be reached with 9 iterations. With p = 0.088, you can get the best EE in 7 iterations. These results illustrate that the convergence rate is dependent on the value of p .

It also analyzes the convergence between "Dinekelbach based CA", "LA OMA", PA, and BIMPA and shows the performance degradation between EE and the rate of convergence. Compare "BIAPA" with CA, Dinikelbach's "LAOMA". As the iterative index increases, you can see that the EE decreases first and then the system EE remains unchanged. It can be said that "BIAPA", "PA" and "CA from Dinkelbach" show very similar performance. Figure 5 shows that the difference in EE performance is negligible and that the distributed method converges faster.