Thanks for the question. The problem with the threshold is how to obtain its level. What ML does can be understood as adjusting the threshold in an automatic manner. If we consider the real network user throughput depends on many functional blocks like scheduler, precoder, beam management etc. Moreover, the realistic massive MIMO radio channel has some correlations, and complicated reflections. Due to those facts it is hard to estimate the user throughput related to the given array configuration apriori, on the basis of standard optimization. It is necessary to capture the data (user troughputs, power consumption) related to the given antenna array, and train the ML model. Moreover, while considering massive MIMO not only the total number of users matters but their spatial distribution, especially in the angular domain. For example, the same number of users can require different array configuration when being cumulated at one specific direction, and when being spread over multiple directions. Our rApp takes multiple input features, and some of them are related to the per beam load distribution that reflects this spatial diversity of users in cell. We hope this clarifies the ML usage.
Hello, thanks for the question. Yes, that is correct - the ES use case is not yet available in the public version of the Open RAN xApp Demonstrator. We're currently working on it.
@@RimedoLabs available in the public version of the Open RAN . You say ES-rApp is now available in the demonstrator. Is it in public domain this demonstrator?
What is the benefit of using machine learning here?
Could the same dynamic behavior be optain based on the throughput trash hold analysis?
Thanks for the question. The problem with the threshold is how to obtain its level. What ML does can be understood as adjusting the threshold in an automatic manner. If we consider the real network user throughput depends on many functional blocks like scheduler, precoder, beam management etc. Moreover, the realistic massive MIMO radio channel has some correlations, and complicated reflections.
Due to those facts it is hard to estimate the user throughput related to the given array configuration apriori, on the basis of standard optimization. It is necessary to capture the data (user troughputs, power consumption) related to the given antenna array, and train the ML model.
Moreover, while considering massive MIMO not only the total number of users matters but their spatial distribution, especially in the angular domain. For example, the same number of users can require different array configuration when being cumulated at one specific direction, and when being spread over multiple directions.
Our rApp takes multiple input features, and some of them are related to the per beam load distribution that reflects this spatial diversity of users in cell.
We hope this clarifies the ML usage.
@@RimedoLabs yes, thank you for the time :).
Hello. Thank you for nice Demo. Is it right that this rApp service is not available on Open RAN xApp demonstrator now? I can only use BMM xApp.
Hello, thanks for the question.
Yes, that is correct - the ES use case is not yet available in the public version of the Open RAN xApp Demonstrator. We're currently working on it.
@@RimedoLabs Is there a prediction for it to be available?
Hello, now the ES-rApp is available in the demonstrator.
@@sergiomartins4652 Could you please clarify the question. What do you mean by "available"?
@@RimedoLabs available in the public version of the Open RAN . You say ES-rApp is now available in the demonstrator. Is it in public domain this demonstrator?