Question Regarding Abnormally High SNR Calculation in Sionna Ray-Tracing Channel

[hssam1002]

Hi, all.
Thanks for your valuable work and for sharing it with the community.

I was a wireless engineer for 7 years, and I'm currently having a hard time with an issue in the RT simulator where I'm seeing SNR values that are much larger than anything I've ever encountered in the field. I would appreciate some guidance on this.

1. System Configuration & SISO SNR Calculation

First, I've assumed the following system parameters:

• Tx Power ($P_{tx}$): 30dBm (1W)
• Subcarrier Spacing: 30 kHz
• FFT Size: 256
  Based on this, the transmit power per subcarrier is 1 W / 256, which is approximately 5.9 dBm.

For the noise power, assuming a receiver Noise Figure of 7 dB, I calculated it as follows:

• $P_N$ (dBm) = -174 + 10*log₁₀(30,000) + 7 ≈ -122 dBm

My formula for the final SNR per subcarrier is:

• SNR (dB) = $P_{rx}$ - $P_N$ = (5.9 dBm - Path Loss) - (-122 dBm) = 127.9 dB - Path Loss

I calculated Path loss term from the CIR provided by path_solver. Specifically, I calculated the path loss as
$P_{loss} = - \sum_{\ell =0}^{L-1} \left| H[\ell] \right|^2_F / N_{FFT}$
However, using this method, the resulting SNR is much higher than I expected (usually exceeding 40 dB), even when LOS is disabled.

2. Extension to Massive MIMO & Observed Issues

A similar issue occurred when I extended this to a Massive MIMO environment.
I assumed the precoder, F, is normalized to have $\text{Tr}\left(F[k]F^*[k]\right) = N_T$ (number of transmit antennas).

To normalize the noise variance to 1, I multiplied the CIR by a factor like $\frac{1}{\sqrt{P_N N_T N_{FFT}}}$.
Thus
$H[\ell] \leftarrow H[\ell] / \sqrt{{P_N N_T N_{FFT}}}$

Just like in the SISO case, the SNR remained abnormally high even before applying any precoding. This caused numerical instability during LMMSE channel estimation, as the covariance matrix Ryy became ill-conditioned, making its inversion problematic.

3. Questions

1. Is there a fundamental misunderstanding in my approach to calculating SNR or normalizing from the RT channel's output (e.g., how I interpret path loss from the CIR)?
2. Is it generally expected for SNR to be this high in ray-tracing simulations, even when the LOS path is disabled?
   In the tutorial at [https://nvlabs.github.io/sionna/rt/tutorials/Radio-Maps.html], the SINR from a single transmitter seems unrealistically high.
3. What is a typical Base Station (BS) transmit power (in dBm) that is commonly used in academic or industry simulations for these kinds of setups?

[jhoydis]

Hi @hssam1002,

1. I think that your approach looks correct apart from the way that compute the path loss (it can only be negative in the logarithmic domain, but I guess that the "-" sign you have is just a typo). We typically compute the path gain from the CIR as:

$$ g = \sum_i |a_i|^2 $$

The path loss would then be $P_\text{loss} = -10\log_{10}(g)$ (note that this is a positive number). For example, in free space at a distance of 100m and carrier frequency of 2.5GHz assuming isotropic antennas, this would be $P_\text{loss}=80.4\text{dB}$ (according to Frii's formula).
In your example above, this would lead to an SNR of 47.5dB. So I do not know what numbers you are expecting. At 1km distance, the path loss would be around 100dB and the SNR around 28dB. Also, 1W TX power for a bandwidth of 7.68MHz is probably a lot. If you would assume 100MHz bandwidth, your SNR would immediately decrease significantly.

2. It is not ray-tracing simulations, it's physics. If you stay closed to a base station with good channel conditions, the SNR can be quite high. What typically limits performance is the interference. If you believe that there is a bug in the way Sionna RT computes the SNR, please open an issue with a code snippet that reproduces unexpected behavior. I do not think that this is the case in the tutorial notebook you mentioned.

Femtocell (residential)	< 0.1 W (often <0.02 W)	< 20 dBm
Picocell (enterprise)	0.02 W – 0.25 W	13–24 dBm
Metrocell (urban)	0.25 W – 6 W	24–38 dBm
Rural small cell	up to 10 W	up to 40 dBm

According to this doc.

[hssam1002]

Thanks for the clear explanation.
That's a great point that the main limiting factor for SINR is interference from the other transmitters; I hadn't considered that before.

I am setting up my simulation follow the Sionna tutorial,
https://nvlabs.github.io/sionna/phy/tutorials/Link_Level_Simulations_with_RT.html
, which covers a range of 5m ~ 400m. From this, I understand the scenario can be considered as a "Metrocell".

So, if I define the transmit power as 24 dBm and use an FFT size of 4096, I can expect the SNR will decrease by about 15dB compared to my previous assumptions.!

Thanks for the help!