PEACH: Proactive and Environment Aware Channel State Information Prediction with Depth Images

Up-to-date and accurate prediction of Channel State Information (CSI) is of paramount importance in Ultra-Reliable Low-Latency Communications (URLLC), specifically in dynamic environments where unpredictable mobility is inherent. CSI can be meticulously tracked by means of frequent pilot transmissions, which on the downside lead to an increase in metadata (overhead signaling) and latency, which are both detrimental for URLLC. To overcome these issues, in this paper, we take a fundamentally different approach and propose PEACH, a machine learning system which utilizes environmental information with depth images to predict CSI amplitude in beyond 5G systems, without requiring metadata radio resources, such as pilot overheads or any feedback mechanism. PEACH exploits depth images by employing a convolutional neural network to predict the current and the next 100 ms CSI amplitudes. The proposed system is experimentally validated with extensive measurements conducted in an indoor environment. We prove that environmental information can be instrumental towards proactive CSI amplitude acquisition of both static and mobile users on base stations, while completely avoiding the dependency on feedback and pilot transmission for both downlink and uplink CSI information. Furthermore, compared to demodulation reference signal based traditional pilot estimation, in ideal conditions without interference, our experimental results show that PEACH yields the similar performance in terms of average bit error rate. More importantly, in the realistic cases with interference taken into account, our experiments demonstrate considerable improvements introduced by PEACH in terms of normalized mean square error of CSI amplitude estimation when compared to traditional approaches.

Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s

INTRODUCTION
Ultra-Reliable Low-Latency Communications (URLLC) is one of the three service types supported in 5G networks. These services are characterized by very stringent traffic requirements to deliver within the order of milliseconds the vast majority of the packets. Autonomous driving, remote surgery, remote monitoring and control, and industrial automation in general are some use cases that belong to services with URLLC traffic. The aforementioned services are not only sensitive to abiding by those very non-flexible requirements, but also, because of their nature, a failure to comply either with the low-latency or reliability requirement can give rise to a serious risk to human lives. Therefore, the paramount importance of enabling (almost) flawless operation of this type of traffic.
Although 5G networks are being deployed since recently, their shortcomings are already recognized when it comes to supporting URLLC traffic [5]. Moreover, beyond 5G/6G wireless systems are envisioned to pave the road further for industrial automation with even more stringent latency and reliability requirements [9], especially for mission-critical communications [5].
The enhancement of the reliability depends heavily on accurate channel estimation to mitigate wireless channel distortion, which requires channel probing. On the other hand, the reduction of latency requires lower processing times and less signaling. Given that these two requirements are conflicting, accurate channel estimation (metadata increases) and channel information feedback (latency increases) for the URLLC are one of the main challenges [3]. Predicting fading wireless channels correctly is of high importance for reliable communication and is still very challenging [4]. While the standardization for 5G targeted latency reduction via new subcarrier spacing (SCS) structure to have shorter slot duration and allowing shorter transmission time intervals (TTIs), the signaling SIGMETRICS '23 Abstracts, June 19-23, 2023, Orlando, FL, USA.
Serkut Ayvaşık, Fidan Mehmeti, Edwin Babaians, and Wolfgang Kellerer overhead for control messages was initially neglected. However, especially for URLLC where small packets are expected, the metadata such as channel estimation feedback, pilot transmissions, or grantbased random access procedures have to be taken into account [3].
While the signaling component in 5G is much more flexible and reduced compared to LTE, there is still a considerable signaling part required to support URLLC services in 5G, mainly due to channel estimation [5,7]. The metadata for channel estimation is not negligible when considering the short-packet transmission in URLLC, where a higher overhead in terms of resources results in more accurate channel estimation; thus, it yields a better reliability while leaving fewer resources for the actual data which is already limited in the URLLC case. For instance, for bad channel conditions where it is vital for URLLC services to be still operating reliably, the metadata-data ratio is observed to be increasing up to 61% [7]. Moreover, channel estimation metadata, namely demodulation reference signal (DMRS), is only capable of providing channel knowledge on the used bandwidth. However, for channel-dependent scheduling or link adaptation, base stations require channel sounding where the sounding reference signal (SRS) is used. Since SRS is transmitted on a wider bandwidth, it imposes higher energy consumption on the user side and also results in larger radio overhead.
Along with the above UL concerns, there are several issues on the downlink (DL) as well. FDD systems require a CSI feedback mechanism for DL CSI knowledge which consumes UL spectrum resources due to lack of reciprocity. The same channel sounding concern also applies to DL resource allocation, in terms of link adaptation and scheduling, for which the CSI reference signal (CSI-RS) is used in a similar manner. CSI-RS also requires users to consume more energy when communicating on a wider bandwidth. Further, the information obtained via CSI-RS needs to be reported back to base stations. However, imperfect CSI caused by outdated CSI knowledge was shown to be affecting the performance significantly [8].
There are several important questions, both from research and practical perspective, which arise related to the correct prediction of channel conditions while reducing metadata in wireless systems: (1) How to provide correct CSI to base stations without or with few metadata on both UL and DL? (2) How to make intelligent scheduling and link adaptation without channel sounding? (3) How to incorporate environment-specific information towards proactive decision-making in radio resource management (RRM)?
To answer these questions, we address CSI amplitude acquisition without any pilot or feedback mechanisms with look-ahead capability and environmental awareness. We build a system for Proactive and Environment-Aware channel state information prediction (PEACH), where depth images are used to provide CSI amplitudes for the current moment and 100 ms into the future. PEACH leverages a convolutional neural network (CNN) to map depth images of the communication environment to CSI amplitudes. The proposed system is implemented and evaluated on real data obtained from extensive measurements over SDRs in an indoor environment for multiple receivers and multiple transmitters. A typical scenario for such an indoor setting would be an industrial factory floor. The performance of PEACH is compared with the traditional DMRS-based methods. PEACH is able to maintain the same bit error rate (BER) performance in bad channel conditions with more conservative modulation schemes, such as QPSK, while not sacrificing too much on higher modulations in an ideal communication scenario without interference. The system is also tested in a more realistic scenario with interference, where the experimental results exposed a clear advantage of PEACH in terms of better normalized mean squared error (NMSE) performance compared to DMRS (of up to 6 dB). The proposed solution provides knowledge on the entire bandwidth, which makes channel sounding obsolete, and thus the SRS usage on UL or CSI-RS usage in DL can be avoided. Further, PEACH provides CSI knowledge into the future, where knowing the channel for the next transmissions was already shown to benefit the scheduling policies [6].
Our main contributions are: (1) We propose a system solution for channel estimation based on video images that reliably reduces the latency for radio resource management and does not require additional metadata. (2) We design and evaluate an ML framework that elucidates how to benefit from environmental awareness in wireless communications towards proactive and intelligent RRM. (3) To the best of our knowledge, this is the first work that demonstrates the prediction with a high accuracy of per-subcarrier CSI amplitude in OFDM systems, specifically in 5G NR, for multiple receivers and multiple transmitters (including mobile transmitters) by using only camera images. (4) To the best of our knowledge, this is also the first public 5G dataset with real wireless raw signals (I/Q samples) obtained over Software-defined radios (SDRs). For future research, raw wireless 5G datasets along with the camera videos are publicly available at [2]. Please check the full paper in [1] for details.