Questions about beam scanning and the dataset

Hi Sachira,

Thank you for your interest in the DeepSense 6G dataset. Please find the answers to your queries below:

1. Multiple Transmitters and Base Stations: In our published scenarios, we currently do not have any that include more than one transmitter or base station. However, we do offer a distributed scenario with two distributed nodes alongside a transmitter and receiver. Each node is equipped with a GPS receiver and an RGB camera. Please refer to https://www.deepsense6g.net/scenario-41/&source=gmail&ust=1713295698430000&usg=AOvVaw22F-Gpj3nPsqa8y9FibUj p">Scenario 41 for detailed information.

2. Received Signal Power Measurement: The received beam power is determined by averaging the power from a set of samples collected via a USRP. The number of samples can vary significantly depending on the testbed configuration and the processing capabilities. For instance, the V2V testbed averages the power over fewer samples than Testbed 1 due to the increased number of sensors in the testbed. The number of samples acquired per beam varies between 64*50 samples and 64*500 samples. Since the sampling rate of the B210s is always the same (61.44 MSamples/s), the sampling interval varies between 0.5 ms (half of a millisecond) and 0.05 ms. Note that each beam measurement takes less than ~1.5 ms to allow for measuring 64 beams 10x per second, which is why we do not acquire more samples. Note further that this should not make a difference because the average number of samples is usually enough to make a good measurement.

3. Beam Sweeping Technique: We do a sequential beam sweep across each beam in the codebook, following the same order each time. The codebook is an oversampled version of the DFT codebook. The codebook has an oversampling factor of 4 (note that we have 64 beams in the codebook but only 16 antennas) - this is the default codebook that is shipped with the phased arrays. You can find the codebook on the https://www.deepsense6g.net/data-collection/&source=gmail&ust=1713295698430000&usg=AOvVaw17IB7JTyMKHO0C5VZoOv6 O">Data Collection page. 

4. Performance Evaluation Methodology: To identify the user in the wireless environment, we first use additional sensing modalities (e.g., GPS position or mmWave power vectors) and a pre-trained ML model to predict probable transmitter center coordinates. It is important to note that these predictions are just estimates, not actual user coordinates. Next, the YOLOv3 model is employed to determine the bounding box center coordinates of all the objects of interest in the wireless environment. We then calculate the Euclidean distance between these predicted coordinates and those of all detected objects. The object with the minimum distance to the predicted transmitter coordinates is identified as the transmitter. Both predicted, and ground-truth coordinates are considered with a precision of up to four decimal places in this work.

Should you require further information or have any additional questions, please do not hesitate to reach out.

Thanks and Regards, 
Team DeepSense

Thanks for your answering. However, I still have some questions on the received signal power. I want to add noise to this raw data for my research, so I need to know the power of the transmitted symbols, as well as the symbol type, that is being used to calculate the noise power through the signal-to-noise ratio. 

Thanks.

Best Regards

Hi @david,

Hi,

Thank you for the helpful information. I have two related questions:

1. Could you clarify the normalization method used for power in your dataset? Specifically:

- Is the power measured in milliwatts?

- Have you normalized the power to the maximum of all samples in the dataset, or is there another reference point used for normalization?

2. What is the sensitivity, i.e., the minimum detectable signal range of the receiver?

Thank you for your help,

Saba 

Hi @sabamhosseini,

Thanks for your question. And our apologies for our delay in replying. 

The power quantity we include in the datasets is dimensionless. Let me explain. To achieve an absolute power measurement, like in Watts or dBm, the components of software-defined radios require an intermediate calibration step. This step is often quite time-consuming and expensive. Therefore, most SDRs will measure relative power instead, which turns out to be sufficiently good for most problems. This means that if you see power values of 2 and 0.5, they cannot be translated to an absolute scale like Watts, but you know that the power measured as 0.5 is 4x smaller than the power measured as 2. Since our datasets also have the locations of transmitters and receivers, one is able to correlate, for example, the relative power measurement with the path loss. 

This almost fully answers your second question about sensitivity as well. The USRPs we use have a gain control setting that we can adjust to amplify more or less the transmitted and received signals. In V2V scenario 36, we performed mmWave measurements roughly 500 meters away using relatively conservative gains. However, we cannot tell in absolute terms what the sensitivity is. In relative terms, it might not be a very useful quantity either because it varies across scenarios due to different noise levels.

Let me know if this answers your question.

Best,

João