Stream Classification in Call Quality Dashboard (CQD)

מאמר
10/22/2024
חל על:

Microsoft Teams, Skype for Business

The Call Quality Dashboard (CQD) for Microsoft Teams and Skype for Business allows you to gain insights into the quality of calls made using Microsoft Teams and Skype for Business services. This topic provides detailed information about the quality classification of media streams. To learn more about CQD and how to set it up, see Set up Call Quality Dashboard.

For a broader and deeper view of call telemetry that looks at several factors to determine perceived user experience of calls and to identify possible root cause when there's a suspected degradation in quality, see Intelligent media quality classifiers in CQD.

Classifier Definitions

Streams in CQD are classified as Good, Poor, or Unclassified based on the values of the available key quality metrics. The metrics and conditions used to classify stream are shown in the tables that follow. CQD's "Poor Due To" dimensions can be used to understand which metric is responsible for a Poor classification. For more information on these dimensions, see Dimensions and measures available in Call Quality Dashboard.

Audio Classifier

If one or more of the following conditions are met and Packet Utilization is > 500 packets, an audio stream is marked as Poor:

Metric	Scenario	Condition	Explanation
Round Trip	ALL	> 500	Average round-trip network propagation time, computed in milliseconds. Details available in RFC3550.
Packet Loss Rate	ALL	> 0.1	Average packet loss rate for stream.
Jitter	ALL	> 30	Average jitter for stream in milliseconds.

Note

The Audio Classifier uses basic network performance telemetry to assess if the optimal conditions for good quality audio were present. If any one of the thresholds are broken then the audio will be marked Poor, but this doesn't mean the audio stream was actually of poor quality, nor does it mean the user perceived a quality issue. The Teams media stack is built to withstand and correct for poor network conditions, and can mitigate considerable network performance degradation in excess of the thresholds above before a drop in quality is perceived by users. We recommend admins make every effort to build and configure their networks and Teams deployments for the best possible quality, and the above metrics are a useful guideline to assess performance of those networks.

Video Classifier due to Freeze

The video stream is marked Good or Poor based on the value of a classifier score generated to estimate that the end user experienced Frozen Video. This classifier is available for Microsoft Teams product only.

Step #	Metric	Scenario	Condition	Classification if Condition is True	Classification if Condition is False	Classification if Metric is Unavailable	Explanation
1	Video Poor Due to Freeze Classifier	Is Server Pair is Client : Server	>0.246	Poor	Good	Unclassified	A Score between 0 and 1 that is generated based on a combination of user experience, freeze duration statistics, and overall call experience
2	Video Poor Due to Freeze Classifier	Is Server Pair is Client : Client	>0.524	Poor	Good	Unclassified	A Score between 0 and 1 that is generated based on a combination of user experience, freeze duration statistics, and overall call experience

Video Classifier

A video stream is marked as Good or Poor based on the value of the first available metric in the following order:

Step #	Metric	Condition	Classification if Condition is True	Classification if Condition is False	Classification if Metric is Unavailable	Explanation
1	Video Local Frame Loss Percentage Avg	> 50%	Poor	Good	Proceed to step 2	Average percentage of video frames lost as displayed to the user. The average includes frames recovered from network losses.
2	Video Frame Rate Avg	< 7	Poor	Good	Proceed to step 3	Average frames per second received for a video stream, computed over the duration of the session.
3	Video Post FECPLR	> 0.15	Poor	Good	Unclassified	Packet loss rate after FEC has been applied aggregated across all video streams and codecs.

VBSS Classifier

A VBSS stream is marked as Good or Poor based on the value of the first available metric in the following order:

Step #	Metric	Condition	Classification if Condition is True	Classification if Condition is False	Classification if Metric is Unavailable	Explanation
1	Video Local Frame Loss Percentage Avg	Codec is NOT H264S And StreamDirection is Inbound If FrameLoss > 50%	Poor	Good	Unclassified	Average percentage of video frames lost as displayed to the user. The average includes frames recovered from network losses. FrameLoss is only used for classifying inbound non-H264S streams.
2	Video Frame Rate Avg	< 1	Poor	Good	Unclassified	Average frames per second received for a video stream, computed over the duration of the session. Applies to all outbound streams and either StreamDirection for H264S.

An application sharing stream is marked as Poor if one or more of the following conditions are met:

Metric	Condition	Explanation
Spoiled Tile Percent Total	> 36	Percentage of tiles that are discarded instead of sent to a remote peer (for example, from the MCU to a viewer). Discarded (or spoiled) tiles might be caused by bandwidth restrictions between client and server.
AppSharing RDP Tile Processing Latency Average	> 400	Average latency in milliseconds processing tiles on the RDP Stack at the conferencing server.
AppSharing Relative OneWay Average	> 1.75	Average relative one-way delay between the endpoints in seconds for application sharing streams.

Unclassified Streams

In CQD, a stream is marked Unclassified when Interactive Connectivity Establishment (ICE) connectivity fails or when all the metrics required to compute the stream classification aren't reported.

To check for ICE connectivity failures, examine the dimensions "First Connectivity Ice" and "Second Connectivity Ice" for a "FAILED" value. If either value indicates a failure, the stream is marked as Unclassified.

If ICE connectivity succeeded for an Unclassified stream, the stream is likely considered Unclassified because key stream metrics weren't reported. There are a few reasons these metrics might not be reported:

QoE reports weren't received — The metrics used for classification are reported in a QoE report sent at the end of a call. If this report isn't produced (for example, because some third-party endpoints might not send QoE) or couldn't be sent (for example, because of a network outage), CQD is unable to classify the stream.

Tip

The "QoE Record Available" dimension can be used to determine whether a QoE report was received for a stream. Note that this dimension will have a value of "True" if a QoE report was received from either endpoint. A QoE report from both endpoints is required for the most accurate reporting of metrics.
Short calls — Short calls might not have enough media activity to compute key stream metrics. Without these metrics, CQD is unable to classify the stream.

Tip

The dimensions "Duration (Seconds)", "Duration (Minutes)", "Duration 5 seconds or less", and "Duration 60 seconds or more" can be used to determine the duration of a stream. The measurement "Avg Call Duration" can also be used to compute the average duration for a set of streams.
Low packet utilization — Like the "short call" scenario, sufficient packet utilization is required for computation of key stream metrics. Without these metrics, CQD is unable to classify the stream.
- A common low packet utilization scenario occurs when an attendee joins a meeting to listen to the presenter, but never speaks (the microphone is muted for most of the call). Here, the audio stream inbound to the client has high packet utilization while the audio stream outbound from the client has little to no packet utilization. The duration of the stream might be an hour or longer but the packet utilization on the stream from the client to the server is low since the microphone was muted, and an Unclassified stream results.
Tip

The "Packet Utilization" dimension and "Avg Packet Utilization" measurement can be used to determine the packet activity of a stream.