Digital media and its network demands
Voice and video now make up most internet traffic. Understanding how media is digitised and compressed explains what it demands of networks.
Digitising audio
Audio is digitised by sampling the sound wave at regular intervals and quantising each sample to a number of bits. Telephone-quality audio samples 8,000 times per second at 8 bits per sample: 64 kbit/s. CD audio uses 44,100 samples per second, 16 bits, and 2 channels, about 1.41 Mbit/s. The Nyquist theorem requires sampling at least twice the highest frequency to be captured.
Audio compression
Compressing audio exploits human hearing: perceptual codecs discard sounds masked by louder ones or outside audible ranges. MP3 and AAC typically deliver good music at 96 to 256 kbit/s; Opus, standardised in 2012, adapts from about 6 kbit/s for speech to 510 kbit/s for music and is used in WebRTC, WhatsApp, and Discord. Speech codecs for voice calls often use 8 to 32 kbit/s.
Digitising and compressing video
Video is a sequence of images: a 1080p frame has 1920 x 1080 pixels, and at 24 bits per pixel and 30 frames per second, raw video needs about 1.5 Gbit/s. Codecs compress within frames, like JPEG, and between frames, sending only changes predicted by motion vectors. Key frames (I-frames) are self-contained; predicted frames (P- and B-frames) refer to others.
Video codecs
H.264/AVC (2003) remains the most widely supported codec. HEVC/H.265 (2013) halves bit rates for similar quality but faces complex licensing. Google's VP9 and the royalty-free AV1 (2018), from the Alliance for Open Media, are used by YouTube, Netflix, and others. Typical streaming bit rates range from about 1 to 3 Mbit/s for standard definition to 15 to 25 Mbit/s for 4K.
Application categories
Multimedia applications fall into three groups. Streaming stored media, such as films on demand, can buffer seconds ahead, tolerating delay. Conversational voice and video, such as calls and video meetings, need low latency, ideally under about 150 milliseconds one way. Live streaming, such as sports broadcasts, sits in between, with a few seconds of delay usually acceptable.
What media needs from the network
Media is sensitive to throughput, delay, jitter, and loss in different ways. Streaming needs sustained throughput above the video's bit rate; conversational media needs low delay and jitter and can tolerate some loss through concealment; all media suffers when congestion makes throughput drop or delay spike. The best-effort internet guarantees none of these, so applications adapt.
Traffic share
Video dominates traffic: industry reports such as Sandvine's have estimated video at well over half of downstream internet traffic, with Netflix, YouTube, and other streaming and social platforms among the largest sources. This is why content delivery networks, efficient codecs, and adaptive streaming matter so much for network operators.
Worked example
A 2-hour film at 5 Mbit/s requires 5 x 7,200 = 36,000 megabits, which is 4,500 megabytes, about 4.5 GB. Encoding the same quality with AV1 instead of H.264 might reduce the bit rate by roughly 30% to 50%, saving 1.4 to 2.3 GB per viewing; multiplied over millions of viewers, codec choice saves enormous amounts of network capacity.
Key takeaways
Audio is sampled and quantised: telephone audio is 64 kbit/s and CD audio about 1.41 Mbit/s, following the Nyquist rate. Perceptual codecs such as MP3, AAC, and Opus compress audio heavily. Raw 1080p30 video needs about 1.5 Gbit/s; codecs exploit spatial and temporal redundancy using I-, P-, and B-frames. H.264, HEVC, VP9, and AV1 trade efficiency, support, and licensing. Stored streaming tolerates delay, conversational media needs under about 150 ms one way, and live streaming sits between. Media is sensitive to throughput, delay, jitter, and loss, and video makes up most traffic.
Flashcard
Measuring a signal at regular intervals
Check your understanding
Why can a film-streaming service tolerate several seconds of delay while a video call cannot?