Answer to your problems
Machine learning - the best way to keep an eye on your network health
KEEPING THE NETWORK HEALTHY IS PROBABLY THE MOST IMPORTANT TASK TELECOMS HAVE TODAY
As the complexity of the cable network increases, it becomes impossible to manually process all the information and recognize the behavior. Machine learning is the answer.
In a healthy network, the conditions are as follows :
- the values are not too close to the threshold values
- the values are stable over time
- the values for different channels differ slightly
Cable modem (CM), as well as CMTS power, should be within the correct thresholds, preferably with a certain margin (between -15 and 15 dBmV) SNR
Signal-to-noise ratio or noise margin – compares the level of the desired signal with the background noise level (should be 30 dB or more)
Machine learning enables automatic problem recognition
Network patterns can be recognized using a variety of machine learning algorithms. For instance, cable modems data grouping to identify problems for multiple customers leads to easier and faster network optimization.
The goal is to identify similar behavior in three steps:
Step 1: collecting and aggregating time-structured data
Step 2: grouping the data into similar groups based on their behavior pattern over time
Step 3: optimizing groups with similar behavior to improve network performance
Once the data has been collected ... it's time for the magic part
In this case, network device data is collected and processed. For example, key network parameters are collected every half hour (e.g. SNR, CER …). In other words, the more information available, the better the analysis.
As expected, such an analysis collects large amounts of data, which are often difficult to understand.
To address this challenge, we use anomaly detection. What does that mean? Time series decomposition is a statistical task that deconstructs a time series into several components, each representing one of the fundamental categories of patterns. As a result, any anomaly in the time series is discarded to track the actual patterns in the data. Therefore, hierarchical clustering produces a set of nested clusters organized as a hierarchical tree. This method is applied to the rest of the time series decomposition.
Reduce enterprise costs with large clusters
Each cluster consists of a set of correlated variables that are classified based on their behavior over time. For example, if noise is recognized in a cluster, the same optimization method is applied to the entire cluster, as opposed to tracking and optimizing individual elements, one at a time.
The result is savings !!
Our solution provides Telecom with an indication of the exact network element that requires intervention. Instead of initiating multiple field interventions for different customers, telecoms assign a single field intervention to solve the problems of a group of customers sharing the same network segment.
This results in savings in the OPEX costs for field technicians, increased customer experience and loyalty, higher network performance score, and faster fault resolution time.