Advanced video specifications

For more information on the video decode and encode capabilities such as the supported profiles, levels, resolutions, frame rates, bit rates, and limitations, select the following tabs:

​

Feature descriptions

The supported encoder feature descriptions are as follows: B-Frame encode B-Frame uses both the earlier and the future frames as reference data to obtain the highest amount of data compression. The Adreno VPU encodes frames with adaptive B type to obtain the highest possible compression without compromising video quality. Encoder initial QP override Video encoding compresses signal levels by mapping them to discrete values. Quantization is a lossy process, and the levels of quantization govern the quality compared to compression. Encoders start with a default Quantization Parameter (QP) at the beginning. Based on the configured bit rate and scene complexity, encoders determine the right QP value by continuously monitoring the complexity and redundancy across frames. The encoder takes a few seconds to reach a steady state and predict the correct QP value that matches the target bit rate, also known as rate convergence. Hierarchical-P encode In the Hierarchical-P (Hier-P) feature, the encoder organizes the frames into many layers, with frames of one layer referencing frames only from the lower layers. The lowest layer, also known as layer 0 or the base layer, is the only exception. The following image shows the layer encoding pattern: In the image, TL-0 represents the base layer, and the remaining layers represent the enhancement layers. Hier-P improves error resilience and temporal scalability. The Hier-P feature is useful for video telephony (VT) or videoconferencing applications that involve channel errors. Hier-P allows you to control error propagation by selectively dropping the enhancement layers. Slice encode Encoders can compress a frame with an independently decodable Group-of-blocks (GOB), also known as slices. If there is a data loss or corruption, each slice is independently decodable and intends to be a unit of recovery. The advantages of introducing slices in an encoded frame are:

• Encoder ignores a corrupt slice and skips to the next slice, thus restricting the corruption to a part of the frame instead of the entire frame.
• Encoder sizes the slices to fit them within a network packet to help with transmission.
• Encoder retransmits erroneous slices instead of sending the entire frame.
• Applications use slices to reduce latency in real-time communication. Applications transmit and decode slices in parallel, eliminating the need to wait for the entire frame to encode.

Slices also work as resynchronization markers because the decoders can resume from the next slice (marker) when there are bit errors. The H.264 and HEVC encoders support slicing on Qualcomm Linux. The number of bits per slice or the number of macroblocks per slice is a slice boundary. Intra-refresh The intra-refresh feature reduces the channel loss in streaming and casting applications that favor a constant bit rate. The Adreno VPU supports random intra-refresh mode. Video encoder preprocessing Applications can use the Adreno VPU to rotate or flip a YUV frame before encoding it. The Adreno VPU rotates or flips the YUV frame without consuming extra power. Rate control The following table lists the supported rate control algorithms: Table: Supported rate control algorithms Long-term reference (LTR) support Video compression works by eliminating redundancies within the frame (intra-frame) and between the frames (inter-frame). Earlier, the encoded frames that used to serve as a basis to derive future frames were known as reference frames. The following reference frames allow advanced encoding applications to control the way a reference frame is stored and referred:

• Short-term reference (STR): The encoder maintains the recent frames in a reference buffer list from the newest to the oldest. The encoder automatically manages frames using STRs for reference, and deletes them from a stored list when they’re no longer used.
• Long-term reference (LTR): Frames that the application can save, use, and remove. The LTR frames help improve quality and ensure error resiliency in video communication. The maximum number of frames that can be marked with LTR frames depends on the device capability.

LTR frames are useful in error-prone channels. Referring to LTR in error-prone channels reduces the possibility of drift errors due to channel losses. The receiver must confirm that the LTR is received successfully and that it can request a new LTR when an error occurs. The network protocols have checksums that can confirm this behavior, along with the decoder corruption flags. New LTR frames are generated from the sender until the receiver confirms that a successful LTR is received. LTR frames are also useful in videos with scene changes where an LTR with the earlier scene is preserved. If that scene comes back, then the LTR can be used effectively. At the start of a new Group-of-pictures (GOP), the encoder automatically fills the LTR slots, with the first slot (slot number 0) containing an IDR frame. Applications can explicitly send an LTR mark request to mark the LTR frames in the appropriate slots. The following flow diagram shows LTR usage and how to mark and use LTR frames on an H.264 or HEVC encoder: Dynamic encoder properties The Adreno VPU encoder supports dynamic change of properties such as bit rate, frame rate, and sync frame. This support allows the application to change the properties and helps in improved visual experience, video data adjusting to network conditions, and minimizing the loss of data during transmission.