21024 DeepBeam: Coordination-Free mmWave Beam Management with Deep Waveform Learning


5G networks use frequencies in the 24-52 GHz range since there are large chunks of bandwidth that can be allocated to mobile operators to provide high data rates to the mobile users. Signals at such high frequencies, however, do not propagate as far as those in the traditional spectrum used in mobile networks (i.e., below 6 GHz).

A solution to this problem is to focus the transmitted energy in narrow beams, so that the distance the signal can travel increases. This, in turn, introduces a new networking problem: the transmitter and the receiver need to point these narrow beams toward each other, otherwise they would be deaf to ongoing transmissions and would not be able to communicate.

The solution to this usually involves a multi-step procedure in which the transmitter and receiver scan different angular directions until they find the other endpoint, exchanging control signaling. The literature and the standards (e.g., 3GPP NR, IEEE 802.11ad/ay) have identified several methods for doing this however they all require some level of coordination between the transmitter and the receiver, and the usage of specific signals (pilots) from the transmitter side. This introduces delays (for example, to establish the link the first time the transmitter and receiver communicate, or to update the pointing directions when one of the two endpoints moves) and overhead (because fewer resources are allocated for data transmissions).

Technology Overview

The method being developed eliminates the use of pilots and the explicit coordination between the transmitter and the receiver. In this development, called DeepBeam, the receiver passively scans data transmissions to other users in the network (without overhead) and learns (using a deep neural network) the set of beams used by the transmitter, associating a quality metric to each of these beams. At the same time, it learns which is the direction of the transmitter with respect to the receiver. Using these two pieces of information, the receiver selects the best pairs of beams (one for the transmitter, one for the receiver) to be used for the communications. By doing this, DeepBeam reduces the latency and the overhead of such procedure. Additionally, it relies on low-level signals from the antenna arrays, which do not need to be processed by the protocol stack of the devices. This makes DeepBeam versatile and deployable with any mmWave networking standard. 

Because of their highly directional transmissions, radios operating at millimeter wave (mmWave) frequencies need to perform beam management to establish and maintain reliable mmWave links. To achieve this objective, existing solutions mostly rely on explicit coordination between the transmitter (TX) and the receiver (RX), which significantly reduces the airtime available for communication and further complicates the network protocol design.


  • This framework for beam management in mmWave networks that does not require pilot sequences from the transmitter, nor any beam sweeping or synchronization from the receiver because the  different beam patterns introduce different “impairments” to the waveform, which can be subsequently learned by a convolutional neural network (CNN).
  • Uses deep learning to identify transmitted beams and the angle of arrival of a millimeter wave signal, without explicit coordination or pilots
  • Accelerates the beam management process independently on the protocol stack of the device
  • Applies deep learning to raw I/Q samples at mmWave frequencies
  • Protocol-stack-independent framework
  • DeepBeam achieves accuracy of up to 96%, 84% and 77% for the classification of a 5-beam, 12-beam and 24-beam codebook
  • Reduces latency by up to 7x with respect to the 5G NR initial beam sweep in a default


  • Use in ad hoc scenarios, where most of prior work assumes the usage of external information to perform beam management 
  • Beam management for 5G and beyond cellular networks operating at mmWaves
  • Beam management in ad hoc and vehicular networks at mmWaves
  • Passive eavesdropping and classification of mmWave radio signals
Patent Information:
For Information, Contact:
Mark Saulich
Associate Director of Commercialization
Northeastern University
Michele Polese
Francesco Restuccia
Tommaso Melodia
5G Networks
5g NR
Artificial intelligence
Beam Management
Deep learning
Machine Learning
Millimeter Wave Spectrum
Mobile devices
Next-generation Mobile Networks