Can AI Shape Cellular Network Operations?
Tue, 01 Dec 2020 17:55:41 -0000
|Read Time: 0 minutes
Mobile network operators (MNOs) are in the process of overlaying their conventional macro cellular networks with shorter-range cells such as outdoor pico-cells. This substantially increases network complexity, which makes OPEX planning and management challenging. Artificial intelligence (AI) offers the potential for MNOs to operate their networks in a more cost-efficient manner. Even though AI deployment has its challenges, most agree such deployment will ease emerging network, model, and algorithm complexity.
Advancements in error coding and communication design have resulted in the performance of the point-to-point link being close to the Shannon limit. This has proven to be effective for designing the fourth generation (4G) long-term evolution (LTE)-Advanced air interface, which has multiple parallel point-to-point links. However, 5G air interfaces are more complicated due to their complex network topology and coordination schemes, and vastly diverse end-user applications. Deriving any performance optimum is computationally infeasible. AI, however, can tame the network complexity by providing competitive performances.
Cellular networks have been designed with the goal of approximating end-to-end system behavior using simple modeling approaches that are amenable to clean mathematical analysis. For example, practical systems use digital pre-distortion to linearize the end-to-end model, for which information theory provides a simple closed-form capacity expression. However, with non-linearities in the wireless channel (e.g., mm-Wave) or device components (e.g., power amplifier), it’s difficult to analytically model such behaviors.
In contrast, AI-based detection strategies can easily model such non-linearities. There are examples in cellular networks where the optimal algorithms are well characterized but complex to implement in practice. For example, for a point-to-point multiple-input-multiple-output (MIMO) link operating with an M-ary quadrature amplitude modulation (QAM) constellation and K spatial streams or reconstruction in compressive spectrum sensing, optimum solutions are extremely complex. In practice, most MIMO systems employ linear receivers, e.g., linear minimum mean squared error (MMSE) receivers, which are known to be sub-optimal yet are easy to implement. AI can offer an attractive performance–complexity trade-off. For example, a deep-learning-based MIMO receiver can provide better performance than linear receivers in a variety of scenarios, while retaining low complexity.
Deep learning can be used for devising computationally efficient approaches for physical (PHY) layer communication receivers. Supervised learning can be used for MIMO symbol detection and channel decoding, fetching potentially superior performance; recurrent neural network (RNN)-based detection can be used for MIMO orthogonal frequency division multiplexing (OFDM) systems; convolutional neural network (CNN)-based supervised learning techniques can deliver channel estimation; unsupervised learning approaches can be used for automatic fault detection and root cause analysis leveraging self-organizing maps; deep reinforced learning (DRL) can be used for designing spectrum access, scheduling radio resources, and cell-sectorization. An AI-managed edge or data center can consider diverse network parameters and KPIs for optimizing on-off operation of servers while ensuring uninterrupted services for the clients. Leveraging historical data collected by data center servers, it’s possible to learn emerging service-usage patterns.
Standards bodies like the Third Generation Partnership Project (3GPP) have defined Network Data Analytics Function (NWDAF) specifications for data collection and analytics in automated cellular networks (3GPP TR 23.791 specification). By leaving AI model development to implementation, 3GPP provides adequate flexibility for network vendors to deploy AI-enabled use cases. The inbound interfaces ingest data from various sources such as operation, administration, and maintenance (OAM), network function (NF), application function (AF), and data repositories; the outbound interfaces relay the algorithmic decisions to the NF and AF blocks, respectively.
In addition to 3GPP, MNOs (AT&T, China Mobile, Deutsche Telekom, NTT DOCOMO, and Orange) established the O-RAN Alliance (https://www.o-ran.org/) with the intent to automate network functions and reduce operating expenses. The O-RAN architecture, which is shown in the following figure, includes an AI-enabled RAN intelligent controller (RIC) for both non-real time (non-RT) and near-real time (near-RT), multi-radio access technology protocol stacks.
Figure: O-RAN Architecture (source: O-RAN Alliance)
The non-RT functions include service and policy management, higher-layer procedure optimization, and model training for the near-RT RAN functionality. The near-RT RIC is compatible with legacy radio resource management and enhances challenging operational functions such as seamless handover control, Quality of Service (QoS) management, and connectivity management with AI. The O-RAN alliance has set up two work groups standardizing the A1 interface (between non-RT RIC and near-RT RIC) and E2 interface (between near-RT RIC and digital unit [DU] stack).
Even though AI shows great promise for cellular networks, significant challenges remain:
- From a PHY and MAC layer perspective, training a cellular AI model using over-the-air feedback to update layer weights based on the back-propagation algorithm is expensive in terms of uplink control overhead.
- Separation of information across network protocol layers make it difficult to obtain labeled training data. For example, training an AI model residing within a base-station scheduler might be challenging if it requires access to application layer information.
- It is important for cellular networks to be able to predict the worst-case behavior. This isn’t always easy for non-linear AI building blocks.
- Cellular networks and wireless standards have been designed based on theoretical analysis, channel measurements, and human intuition. This approach allows domain experts to run computer simulations to validate communication system building blocks. AI tools remain black boxes. It is still challenging to develop analytical models to test correctness and explain behaviors in a simple manner.
- If a communication task is performed using an AI model, it is often unclear whether the dataset used for training the model is general enough to capture the distribution of inputs as encountered in reality. For example, if a neural network-based symbol detector is trained under one modulation and coding scheme (MCS), it is unclear how the system would perform for a different MCS level. This is important because if the MCS is changing adaptively due to mobility and channel fading, there has to be a way of predicting system behavior.
- Interoperability is crucial in today’s software defined everything (SDE). Inconsistency among AI-based modules from different vendors can potentially deteriorate overall network performance. For example, some actions (e.g., setting handover threshold) taken by an AI-based module from one vendor could counteract the actions taken by another network module (which may or may not be AI-based) from a second vendor. This could lead to unwanted handover occurrences between the original BS and the neighboring BS, causing increased signaling overhead.
In summary, MNOs agree that:
- Training needs to be distributed as more complex scenarios arise.
- More tools explaining AI decision making are essential.
- More tools are needed to compare AI model output to theoretical performance bounds.
- AI models need to adapt based on surrounding contextual information.
- AI deployment should first focus on wider timescale models until a point is reached when model decision making is indistinguishable from experts.
- Fail-safe wrappers around models should limit impact of cascading errors.
AI can revitalize wireless communications. There are challenges to overcome, but, done right, there is opportunity to deliver massive-scale autonomics in cellular networks that support ultra-reliable low-latency communications, enhanced mobile broadband, and massive machine-to-machine communications.