Recurrent Neural Networks For Accurate RSSI Indoor Localization (1903.11703v2)

Abstract: This paper proposes recurrent neuron networks (RNNs) for a fingerprinting indoor localization using WiFi. Instead of locating user's position one at a time as in the cases of conventional algorithms, our RNN solution aims at trajectory positioning and takes into account the relation among the received signal strength indicator (RSSI) measurements in a trajectory. Furthermore, a weighted average filter is proposed for both input RSSI data and sequential output locations to enhance the accuracy among the temporal fluctuations of RSSI. The results using different types of RNN including vanilla RNN, long short-term memory (LSTM), gated recurrent unit (GRU) and bidirectional LSTM (BiLSTM) are presented. On-site experiments demonstrate that the proposed structure achieves an average localization error of $0.75$ m with $80\%$ of the errors under $1$ m, which outperforms the conventional KNN algorithms and probabilistic algorithms by approximately $30\%$ under the same test environment.

• The paper demonstrates that Recurrent Neural Networks (RNNs), particularly LSTMs, significantly improve WiFi RSSI indoor localization accuracy by processing sequential data and exploiting temporal correlations.
• The proposed LSTM model achieved an average localization error of 0.75 meters, with 80% of errors under 1 meter, outperforming conventional methods by approximately 30%.
• This RNN-based approach offers potential for precise indoor localization in practical applications like smart buildings and can be further improved by incorporating multi-modal sensor data.

Recurrent Neural Networks for Accurate RSSI Indoor Localization

Indoor localization systems using WiFi signals have been steadily evolving but continue to face challenges related to signal fluctuation, spatial ambiguity, and short data collection times. The paper "Recurrent Neural Networks For Accurate RSSI Indoor Localization" addresses these issues by exploiting the sequential nature of user movements and employing recurrent neural networks (RNNs)—specifically LSTM (Long Short-Term Memory) networks—to process sequences of RSSI (Received Signal Strength Indicator) data for enhanced trajectory-based localization accuracy.

Methodology

The research presents an innovative approach by proposing RNNs for WiFi fingerprinting indoor localization. Unlike conventional methods that estimate positions independently, this RNN solution considers the user's movement as a sequence, thus exploiting the temporal correlations in RSSI readings. A weighted average filter is applied to both the input RSSI data and the output locations to mitigate RSSI instability due to environmental changes and multipath effects.

Various RNN architectures, including vanilla RNN, LSTM, GRU, and their bidirectional variants, were evaluated for their ability to predict a user's trajectory by processing sequences of RSSI measurements. These networks aim to improve localization accuracy by integrating "memory" of past signals into the localization process.

Experiments and Results

The paper presents an extensive set of experiments conducted in a detailed test environment. The proposed LSTM model, configured as a P-MIMO (Predictive Multiple Input Multiple Output) system, achieved an impressive average localization error of 0.75 meters, with 80% of localization errors falling under 1 meter. This performance surpasses both conventional machine learning methods such as MLP (Multilayer Perceptron), MLNN (Multi-Layer Neural Network), and traditional deterministic methods like KNN (K Nearest Neighbors) and Kalman filters, reducing errors by roughly 30%.

Through analysis, the paper highlights the robustness of LSTM networks in handling WiFi RSSI's inherent fluctuations and spatial ambiguities. Leveraging a probabilistic mapping model to train the RNN on random trajectories allowed the network to effectively differentiate between ambiguous and actual locations even when RSSI readings were unstable.

Implications and Future Directions

This research contributes significantly to indoor localization advancements by demonstrating that RNNs, particularly LSTMs, are more capable of dealing with the nonlinearities and uncertainties involved in handling RSSI data compared to conventional models. Practically, such systems could be invaluable in smart buildings, warehouses, and other scenarios where precise indoor localization is crucial.

Theoretically, the approach provides insight into how neural networks can be trained on simulated trajectory data to capture complex human navigation patterns, potentially opening avenues for further refinement and use in other forms of sensor-based localization.

Further exploration could focus on reducing the training time and computational demands through modifications like LSTM with a projection layer, which offers parameter reduction capabilities. Additionally, incorporating multi-modal sensor data (e.g., combining RSSI with other sensor inputs) may enhance robustness and accuracy, reducing potential errors linked to RSSI's signal variability further.

In conclusion, this paper provides a comprehensive analysis and application of RNNs for WiFi-based indoor localization, demonstrating their potential beyond conventional, single-point estimation techniques. This methodology paves the way for more adaptive and precise localization systems, capable of improving situational awareness in complex indoor environments.