Qualcomm Explores AI Techniques That Could Improve the Wireless Experience

AI Research at Qualcomm Technologies advances wireless communications and RF location detection.

Wireless communications and RF sensing technologies continue to advance with 5G rolling out now and 5G Advanced in the wings. To maintain high quality of service and deliver a superior user experience, wireless communications equipment makers must anticipate a rapidly changing physical landscape around the transmitter and receiver, while also being able to pinpoint a mobile user’s location beyond the range of satellite services. . A 5G modem in a handset must be able to function in wireless communication channels with obstacles and reflections from objects such as floors and walls. Failure to navigate this complex world effectively can result in dropped calls or poorer signal quality.

A pioneer of advanced wireless technologies and a leader in energy-efficient artificial intelligence (AI), Qualcomm has researched the intersection of these technologies. As an example, at this year’s Mobile World Congress in Barcelona, ​​the research team showed an end-to-end over-the-air (OTA) testbed for an AI-enabled air interface. Machine Learning (ML) and wireless communication make up quite a powerful technology because they have complementary strengths. ML can take wireless communication to the next level, providing not only fast and flexible models and algorithms for communication in a dynamic environment, but also accurate RF-based detection of a mobile user’s environment and location, even in a complex indoor environment such as an office building. Let’s take a look at what Qualcomm achieves by merging these two technologies (check out their webinar for more details).

Using AI to improve wireless communication

First off, we need to be clear that AI is already impacting wireless communications in the Snapdragon X70 Modem-RF system, which will be released later this year. Announced in February 2022, the X70 is the first standalone modem with a built-in AI engine. (The X70 is expected to be used by high-end handset manufacturers.) However, this modem is just the beginning of this journey; Qualcomm Technologies is currently exploring more than a dozen research areas for 5G enhancements with AI, including channel status feedback and mmWave beam management. AI could potentially impact many other 5G-related research areas, including energy conservation, security, contextual awareness, and positioning.

One of the key use cases for applying ML to wireless is using generative modeling to provide a more accurate channel representation and thereby design better algorithms and improve communications. Wireless signals are, of course, rarely transmitted along a straight line of sight, but can be blocked or reflected from multiple surfaces, such as walls, floor, ceiling and objects. Classic channel models that deal with these factors work, but require cumbersome field measurements, have hard-coded model assumptions, and are slow to prototype.

Qualcomm AI Research has shown that neural channel models can accurately match complex data distributions with rapid prototyping. Using a Generative Adversarial Network, or GAN, Qualcomm AI Research has designed a neural channel model that learns to generate multiple antennas (MIMO). The generator learns the channel distribution along with a discriminator that teaches the generator to capture the most relevant wireless functions in the model. The neural channel model is interpretable and can be used to model channels with different configurations. More accurate channel models are essential for better communication design and crucial for evaluating the benefits of AI compared to classical methods.

Another example of applying AI to wireless is improving communication design. Qualcomm AI Research applies neural augmentation to enhance Kalman filters, providing more accurate channel acquisition and improving signal quality. Neural augmentation is a high-level design principle that suggests that classical algorithms designed on the basis of domain knowledge can be improved by integrating machine learning algorithms to tune and adjust algorithm parameters. Communication channels are difficult to estimate accurately because they vary over time with unknown dynamics, and the pilot signal observations are noise. More accurate channel estimation at all time steps for different dynamics allows for more efficient communication. Classic Kalman filters lose accuracy over different dynamics and require adjustment of their parameters to channel dynamics. A neural augmentation of the Kalman filter directly adjusts the Kalman parameters based on the channel dynamics. It shows superior results over both classical Kalman and standalone LSTM ML approaches. By preserving the Kalman filter as the backbone of the model, neural augmentation of Kalman filters provides robust generalization to unseen scenarios, better than a single Kalman, while preserving the interpretability of the model. The performance improvement related to stand-alone Kalman filters comes from the expressive power of neural networks that allow adapting Kalman parameters to more complicated dynamics, e.g. for high Doppler.

Using AI to enable accurate location detection with RF

Location-based services can be challenging when the global navigation satellite system (GNSS) signals are weak or unavailable. Consequently, positioning via AI-enhanced RF sensing indoors and in locations with no clear line of sight to GNSS can be useful for use cases such as indoor navigation, vehicle navigation, AGV tracking and asset tracking.

Two types of RF detection based positioning can be distinguished. In the active In this case, a mobile device actively tries to position itself by transmitting RF signals and measuring the channel. In the passive In this case, RF signals are used as radar to detect objects or persons that do not emit RF signals themselves.

Current active positioning methods have limitations. The classical method of time difference of arrival (TDOA) does not require labeled data, but it suffers from accuracy issues under non-line-of-sight and may not utilize valuable multi-path information. The machine learning-assisted RF fingerprints are more accurate, but require the collection of densely sampled positioning labels upon implementation. Data collection may need to be repeated if the channel changes significantly.

Qualcomm AI Research has developed an ML-based active positioning approach called Neural RF SLAM that achieves the best of both worlds. Neural RF SLAM trains a neural network to predict the location of the 3D device based on channel status information. The training target is based on being able to reconstruct the channel status measurement from the predicted 3D location. The reconstruction of the channel status is done by a tunable RF propagation model. The parameters of the reproductive model are set in addition to the neural network weights during training. Multipath information is used in both mapping and reconstruction. Neural RF SLAM achieves 43.4cm accuracy 90% of the time with just a single access point with a single antenna as the anchor.

Similarly, passive positioning via RF sensing, which can position people only with access points, may soon be widely available. Passive positioning and sensing can be used for various usage situations, such as detecting presence, counting people and monitoring sleep. Qualcomm AI Research’s method, WiCluster, is a weakly controlled ML technique that works well for positioning people in non-line-of-sight environments, such as across floors of buildings. It is weakly supervised in that it only takes a few room-level labels and a floor plan. WiCluster discovers the 3D manifold that represents the subject’s movement in the channel state and links it to the floor plan. During extensive testing, WiCluster has demonstrated accurate subject positioning with errors in the 1-2 m range, results generally comparable to supervised training models requiring expensive data labels.

conclusions

While applying AI to wireless still presents challenges to be addressed, such as generalization beyond the modeled domain, adaptability, and the impracticalities of supervised learning, neurally-enhanced wireless communications and RF sensing hold promise in improving customer experiences. Combined with Qualcomm’s “AI Firsts” AnnouncementsThese recent revelations about combining AI and 5G to improve communications demonstrate Qualcomm Technologies’ unique advantage in modems, application processors and AI research.

Leave a Comment

Your email address will not be published.