# Channel State Estimation in LTE-Based Heterogenous Networks Using Deep Learning

^{*}

## Abstract

**:**

## 1. Introduction

## 2. Related Work

## 3. Data Allocation Algorithm

Algorithm 1 Pseudocode of the data allocation algorithm. |

**Require:**- Transmission requirements;
**Ensure:**- Estimates of transmission effectiveness (BLER); Parameters of N radio interfaces;
whiletruedoState estimation in N radio interfaces before the transmission; Update and predict channel state metric of N radio interfaces; if Transmission request then Select the radio interface; Realize transmission; Estimate the transmission effectiveness; Update the historical metrics of the radio channel estimates; Update the historical metrics of the transmission parameters in N radio interfaces; end ifend while |

Algorithm 2 Pseudocode of the decomposed data allocation algorithm. |

**Require:**- Number of the radio interface;
**Ensure:**- Estimates of transmission effectiveness (BLER); Parameters of the radio interface;
Radio channel state estimation before the transmission; whiletruedo Update channel state metrics of the radio interface; Predict channel state metrics of the radio interface; Estimate the transmission effectiveness; Evaluate the transmission effectiveness; Update the historical metric of the radio channel estimates; Update the historical metric of the transmission parameters of the radio interface; end while |

## 4. Measurement Stand

## 5. Measurement Scenarios

- BLER [%];
- RSRP before and during transmission;
- RSRQ (Reference Signal Received Quality) before and during transmission;
- RSSI before and during transmission;
- SINR before and during transmission;
- MCS (Modulation and Coding Scheme) before and during transmission.

## 6. Experimental Studies

#### 6.1. Measurement Campaign No. 1

#### 6.2. Measurement Campaign No. 2

## 7. Proposed Deep Learning Approach

- number of the input nodes;
- number of the hidden layers;
- number of nodes in the hidden layers (assuming a constant number of nodes in all layers);
- the form of the activation functions;
- learning rate.

- preliminary grid search analysis and potentially coarse determination of the prediction efficiency dependence of the BLER as a function of the network architecture variable parameters;
- extended grid search analysis.

## 8. Conclusions

## Author Contributions

## Funding

## Data Availability Statement

## Acknowledgments

## Conflicts of Interest

## References

- Dorner, S.; Cammerer, S.; Hoydis, J.; Ten Brink, S. On Deep Learning-Based Communication Over the Air. IEEE J. Sel. Top. Signal Process.
**2017**, 12, 132–143. [Google Scholar] [CrossRef] [Green Version] - Moiseev, S.N.; Filin, S.A.; Kondakov, M.S.; Garmonov, A.V.; Savinkov, A.Y.; Park, Y.S.; Yim, D.H.; Lee, J.H.; Cheon, S.H.; Han, K.T. Practical propagation channel robust BLER estimation in the OFDM/TDMA and OFDMA broadband wireless access networks. IEEE Veh. Technol. Conf.
**2006**, 1127–1131. [Google Scholar] [CrossRef] - Avidor, D.; Hartmann, C.; Mukherjee, S. On the uplink SIR of a mobile in soft handoff with applications to BLER estimation. IEEE Trans. Wirel. Commun.
**2005**, 4, 467–474. [Google Scholar] [CrossRef] - Jorke, P.; Guldenring, J.; Bocker, S.; Wietfeld, C. Coverage and Link Quality Improvement of Cellular IoT Networks with Multi-Operator and Multi-Link Strategies. In Proceedings of the IEEE 89th Vehicular Technology Conference (VTC2019-Spring), IEEE, Kuala Lumpur, Malaysia, 28 April–1 May 2019. [Google Scholar]
- Abdullah, E.; Hidayat, N.M.; Aziz, I.A.A.; Shuhaimi, N.I.; Sifuddin, M.A. Comparison of F-OFDM and FBMC in terms of PAPR, BER and FER in Multilink Scenario. In Proceedings of the ISCAIE 2021—IEEE 11th Symposium on Computer Applications and Industrial Electronics, Penang, Malaysia, 3–4 April 2021; pp. 378–382. [Google Scholar] [CrossRef]
- Kántor, P.; Bitó, J. Adaptive link assigment applied in case of video streaming in a multilink environment. In Proceedings of the IEEE Vehicular Technology Conference, San Francisco, CA, USA, 5–8 September 2011. [Google Scholar] [CrossRef]
- Saffar, I.; Morel, M.L.A.; Singh, K.D.; Viho, C. Deep learning based speed profiling for mobile users in 5G cellular networks. In Proceedings of the 2019 IEEE Global Communications Conference, GLOBECOM 2019, Big Island, HI, USA, 9–13 December 2019. [Google Scholar] [CrossRef]
- Lee Loong, Y.; Quin, D. A Survey on Applications of Deep Reinforcement Learning in Resource Management for 5G Heterogeneous Networks. In Proceedings of the APSIPA Annual Summit and Conference, Lanzhou, China, 18–21 November 2019. [Google Scholar]
- Yu, P.; Yang, M.; Xiong, A.; Ding, Y.; Li, W.; Qiu, X.; Meng, L.; Kadoch, M.; Cheriet, M. Intelligent-Driven Green Resource Allocation for Industrial Internet of Things in 5G Heterogeneous Networks. IEEE Trans. Ind. Inform.
**2020**, 18, 520–530. [Google Scholar] [CrossRef] - Imran, M.; Yin, H.; Chen, T.; Huang, Z.; Zhang, X.; Zheng, K. DDHH: A decentralized deep learning framework for large-scale heterogeneous networks. In Proceedings of the 2021 IEEE 37th International Conference on Data Engineering (ICDE), Chania, Greece, 19–22 April 2021; Volume 2021, pp. 2033–2038. [Google Scholar] [CrossRef]
- Zhang, C.; Patras, P.; Haddadi, H. Deep Learning in Mobile and Wireless Networking: A Survey. IEEE Commun. Surv. Tutor.
**2019**, 21, 2224–2287. [Google Scholar] [CrossRef] [Green Version] - Zhang, Y.; Xiong, L.; Yu, J. Deep learning based user association in heterogeneous wireless networks. IEEE Access
**2020**, 8, 197439–197447. [Google Scholar] [CrossRef] - Tang, F.; Zhou, Y.; Kato, N. Deep Reinforcement Learning for Dynamic Uplink/Downlink Resource Allocation in High Mobility 5G HetNet. IEEE J. Sel. Areas Commun.
**2020**, 38, 2773–2782. [Google Scholar] [CrossRef] - Mismar, F.B.; Evans, B.L. Deep Learning in Downlink Coordinated Multipoint in New Radio Heterogeneous Networks. IEEE Wirel. Commun. Lett.
**2019**, 8, 1040–1043. [Google Scholar] [CrossRef] [Green Version] - Basu, S.S.; Sultania, A.K.; Famaey, J.; Hoebeke, J. Experimental Performance Evaluation of NB-IoT. In Proceedings of the International Conference on Wireless and Mobile Computing, Networking and Communications (WiMob), Barcelona, Spain, 21–23 October 2019. [Google Scholar]
- Afroz, F.; Subramanian, R.; Heidary, R.; Sandrasegaran, K.; Ahmed, S. SINR, RSRP, RSSI and RSRQ Measurements in Long Term Evolution Networks. Int. J. Wirel. Mob. Netw.
**2015**, 7, 113–123. [Google Scholar] [CrossRef] - Ding, Y.; Kwon, H. Doppler Spread Estimation for 5G NR with Supervised Learning. In Proceedings of the 2020 IEEE Global Communications Conference, GLOBECOM 2020, Taipei, Taiwan, 7–11 December 2020. [Google Scholar] [CrossRef]
- Singh, I.; Kalyani, S.; Giridhar, K. A practical compressed sensing approach for channel estimation in OFDM systems. IEEE Commun. Lett.
**2015**, 19, 2146–2149. [Google Scholar] [CrossRef] - Fernandez Maimo, L.; Perales Gomez, A.L.; Garcia Clemente, F.J.; Gil Perez, M.; Martinez Perez, G. A Self-Adaptive Deep Learning-Based System for Anomaly Detection in 5G Networks. IEEE Access
**2018**, 6, 7700–7712. [Google Scholar] [CrossRef] - Błaszkiewicz, O.; Burczyk, R.; Czapiewska, A.; Gajewska, M.; Gajewski, S.; Marczak, A.; Miszewski, M.; Sadowski, J.; Stefański, J. Research stand for automatic analysis of transmission quality of the KODEŚ concentrator with LTE, NB-IoT and ISM interfaces (in Polish). Przegla̧d Telekomunikacyjny - Wiadomości Telekomunikacyjne
**2021**, 1, 10–19. [Google Scholar] [CrossRef] - Błaszkiewicz, O.; Burczyk, R.; Czapiewska, A.; Gajewska, M.; Gajewski, S.; Sadowski, J. Transmission quality estimation in LTE basing on indicators measured by radio module (in Polish). XX Krajowa Konferencja Elektroniki
**2021**. [Google Scholar] [CrossRef] - Rohde & Schwarz. CMW500 User Manual. Available online: https://www.rohde-schwarz.com/jp/product/cmw500-productstartpage_63493-10341.html (accessed on 19 November 2021).
- Filliben, J.J. The Probability Plot Correlation Coefficient Test for Normality. Technometrics
**1975**, 17, 111–117. [Google Scholar] [CrossRef] - Pati, P.S.; Sahoo, S.S.; Krishnaswamy, D.; Datta, R. A Novel Machine Learning Approach for Link Adaptation in 5G Wireless Networks. In Proceedings of the 2020 2nd PhD Colloquium on Ethically Driven Innovation and Technology for Society, PhD EDITS 2020, Bangalore, India, 8 November 2020. [Google Scholar] [CrossRef]
- Laskowski, M.; Ambroziak, S.J.; Correia, L.M.; Świder, K. On the usefulness of the generalised additive model for mean path loss estimation in body area networks. IEEE Access
**2020**, 8, 376873–376882. [Google Scholar] [CrossRef] - Goodfellow, I.; Bengio, Y.; Courville, A. Deep Learning; The MIT Press: London, UK, 2016. [Google Scholar]
- Li, G.; Xiao, M.; Guo, Y. Application of Deep Learning in Stock Market Valuation Index Forecasting. In Proceedings of the 2019 IEEE 10th International Conference on Software Engineering and Service Science, Beijing, China, 18–20 October 2019. [Google Scholar]
- Yang, W.; Wang, R.; Wang, B. Detection of anomaly stock price based on time series deep learning models. In Proceedings of the 2020 Management Science Informatization and Economic Innovation Development Conference, MSIEID 2020, Guangzhou, China, 18–20 December 2020; pp. 110–114. [Google Scholar] [CrossRef]
- Zhang, L.; Lu, L.; Wang, X.; Zhu, R.M.; Bagheri, M.; Summers, R.M.; Yao, J. Spatio-Temporal Convolutional LSTMs for Tumor Growth Prediction by Learning 4D Longitudinal Patient Data. IEEE Trans. Med. Imaging
**2020**, 39, 1114–1126. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Xu, L.; Kun, W.; Pengfei, L.; Miaoyu, X. Deep Learning Short-Term Traffic Flow Prediction Based on Lane Changing Behavior Recognition. In Proceedings of the 2020 2nd International Conference on Information Technology and Computer Application, ITCA 2020, Guangzhou, China, 18–20 December 2020; pp. 760–763. [Google Scholar] [CrossRef]
- Patterson, J.; Gibson, A. Deep Learning; O’Reilly Media, Inc.: Newton, MA, USA, 2017. [Google Scholar]
- Cwalina, K.K.; Rajchowski, P.; Blaszkiewicz, O.; Olejniczak, A.; Sadowski, J. Deep learning-based LOS and NLOS identification in wireless body area networks. Sensors
**2019**, 19, 4229. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Kingma, D.P.; Ba, J. Adam: A Method for Stochastic Optimization. In Proceedings of the 3rd International Conference on Learning Representations (ICLR), San Diego, CA, USA, 7–9 May 2015. [Google Scholar]

**Figure 1.**The block diagram of the data allocation algorithm for selecting the optimal radio link in the heterogeneous node.

**Figure 4.**The histogram of the estimated errors of the BLER metric prediction for instantaneous radio parameters.

**Figure 5.**The histogram of the estimated BLER metric prediction errors for prediction using the 15th-order FIR filter.

**Figure 6.**The histogram of the estimated BLER metric prediction errors for prediction using the median for the window length $W=15$.

**Figure 7.**The histogram of the estimated errors of the BLER metric prediction for prediction with the use of separate linear models for each parameter for the window length $W=15$.

**Figure 8.**Matrix of the BLER RMSE metric predictions for the instantaneous values of radio parameters.

**Figure 10.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=10$.

**Figure 11.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=15$.

**Figure 12.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=20$.

**Figure 13.**Matrix of the BLER RMSE metric predictions for the instantaneous values of the radio parameters in the second stage learning process.

**Figure 14.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=5$ in the second stage learning process.

**Figure 15.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=10$ in the second stage learning process.

**Figure 16.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=15$ in the second stage learning process.

**Figure 17.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=20$ in the second stage learning process.

**Figure 18.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=25$ in the second stage learning process.

**Figure 19.**Matrix of the BLER RMSE metric predictions for the radio parameters history window $W=30$ in the second stage learning process.

**Figure 20.**The RMSE BLER as a function of the analyzed time window length (number of the input parameters).

Parameter | Value | |
---|---|---|

Frequency band | OB1 | UL: 1920–1980 MHz |

DL: 2110–2170 MHz | ||

Bandwidth | 10 MHz | |

Fading profile | EP5 Medium | |

Doppler shift | 1 Hz | |

SNR range | from −3 dB to 30 dB every 1 dB | |

Transmitted signal power spectral density | −75 dBm/15 kHz | |

Number of subframes transmitted within a single transmission | 100 | |

CQI (Channel Quality Indicator) | {1, 3, 5, 7, 9, 11} | |

Measurement interval | max. 300 ms |

CQI | ${\mathit{RSRP}}_{\mathit{b}}$ | ${\mathit{RSRQ}}_{\mathit{b}}$ | ${\mathit{RSSI}}_{\mathit{b}}$ | ${\mathit{SINR}}_{\mathit{b}}$ | ||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|

$\mathbf{\mu}$ | $\mathit{M}$ | $\mathit{All}$ | $\mathbf{\mu}$ | $\mathit{M}$ | $\mathit{All}$ | $\mathbf{\mu}$ | $\mathit{M}$ | $\mathit{All}$ | $\mathbf{\mu}$ | $\mathit{M}$ | $\mathit{All}$ | |

1 | −0.4 | −0.4 | −0.4 | −0.4 | −0.4 | −0.5 | −0.4 | −0.4 | −0.3 | −0.4 | −0.4 | −0.4 |

3 | −0.5 | −0.5 | −0.4 | −0.5 | −0.5 | −0.6 | −0.4 | −0.4 | −0.4 | −0.4 | −0.4 | −0.4 |

5 | −0.6 | −0.6 | −0.7 | −0.7 | −0.7 | −0.8 | −0.5 | −0.5 | −0.6 | −0.6 | −0.6 | −0.6 |

7 | −0.7 | −0.7 | −0.8 | −0.7 | −0.7 | −0.9 | −0.6 | −0.6 | −0.7 | −0.6 | −0.6 | −0.8 |

9 | −0.7 | −0.7 | −0.9 | −0.5 | −0.5 | −0.8 | −0.6 | −0.6 | −0.8 | −0.6 | −0.6 | −0.8 |

11 | −0.7 | −0.7 | −0.9 | −0.4 | −0.4 | −0.7 | −0.6 | −0.6 | −0.9 | −0.6 | −0.6 | −0.8 |

**Table 3.**The hyperparameter variability and the learning parameters in the first stage of the learning process.

Parameter | Value |
---|---|

Number of hidden layers | from 1 to 5 |

Number of nodes in the hidden layers | from 20 to 200 with the step of 20 |

Activation function | tansig, ReLU, tanh, sigmoid |

Number of the input nodes | 5 parameters and their multiple depending on the length of the analyzed history, i.e., from 0 to 20; |

Learning rate | 0.001 |

Number of the learning iterations | max. 5000 or max. 20 validation error rate |

Dataset split | 60% learning |

20% validation | |

20% testing |

Number of nodes | 20 | 40 | 60 | 80 | 100 | 120 | 140 | 160 | 180 | 200 |

RMSE [%] | 11.7 | 12.5 | 12.7 | 12.9 | 13.3 | 13.3 | 13.3 | 13.3 | 13.5 | 13.3 |

Number of the history samples | 5 | 10 | 15 | 20 |

RMSE [%] | 11.8 | 11.78 | 12.17 | 11.66 |

Number of the hidden layers | 4 | 4 | 5 | 3 |

**Table 6.**The hyperparameter variability and the learning parameters in the second stage of the learning process.

Parameter | Value |
---|---|

Number of hidden layers | from 1 to 10 |

Number of nodes in the hidden layers | from 20 to 100 with the step of 20 |

Activation function | tansig |

Number of the input nodes | 5 parameters and their multiple depending on the length of the analyzed history, i.e., from 0 to 30; |

Learning rate | 0.001 |

Number of the learning iterations | max. 10,000 or max. 50 validation error rate |

Dataset split | 60% learning |

20% validation | |

20% testing |

**Table 7.**Calculated the smallest RMSE for the selected numbers of the analyzed historical radio parameters.

Number of the historical samples | 5 | 10 | 15 | 20 | 25 | 30 |

RMSE [%] | 11.0 | 10.7 | 11.4 | 11.1 | 11.0 | 11.2 |

Number of the hidden layers | 7 | 4 | 2 | 3 | 3 | 3 |

Number of the nodes | 20 | 20 | 40 | 20 | 20 | 20 |

Parameter | Value |
---|---|

Number of hidden layers | 4 |

Number of nodes in the hidden layers | 20 |

Activation function | tansig |

Number of the input nodes | 55 |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).

## Share and Cite

**MDPI and ACS Style**

Cwalina, K.K.; Rajchowski, P.; Olejniczak, A.; Błaszkiewicz, O.; Burczyk, R.
Channel State Estimation in LTE-Based Heterogenous Networks Using Deep Learning. *Sensors* **2021**, *21*, 7716.
https://doi.org/10.3390/s21227716

**AMA Style**

Cwalina KK, Rajchowski P, Olejniczak A, Błaszkiewicz O, Burczyk R.
Channel State Estimation in LTE-Based Heterogenous Networks Using Deep Learning. *Sensors*. 2021; 21(22):7716.
https://doi.org/10.3390/s21227716

**Chicago/Turabian Style**

Cwalina, Krzysztof K., Piotr Rajchowski, Alicja Olejniczak, Olga Błaszkiewicz, and Robert Burczyk.
2021. "Channel State Estimation in LTE-Based Heterogenous Networks Using Deep Learning" *Sensors* 21, no. 22: 7716.
https://doi.org/10.3390/s21227716