One-Size-Fits-All Wireless Video

Preview:

DESCRIPTION

One-Size-Fits-All Wireless Video. Szymon Jakubczak with Hariharan Rahul and Dina Katabi. Wireless Video Has Important Applications. Mobile TV Live streaming sports, concerts, conferences, lectures, … Broadcast TV. All involve multicast, and some involve mobility - PowerPoint PPT Presentation

Citation preview

One-Size-Fits-All Wireless Video

Szymon Jakubczak

with Hariharan Rahul and Dina Katabi

• Mobile TV• Live streaming– sports,

concerts, conferences, lectures, …

• Broadcast TV

Wireless Video Has Important Applications

All involve multicast, and some involve mobilityCurrent design struggles with multicast and mobility

Multicast Challenges Current Wireless Design

High bitrate Starves the far receiver

6Mb/s

1Mb/s

• Currently, the sender has to pick a bitrate• But different receivers support different bitrates

Multicast Challenges Current Wireless Design

High bitrate Starves the far receiverLow bitrate Reduces everyone to the worst receiver

• Currently, the sender has to pick a bitrate• But different receivers support different bitrates

6Mb/s

1Mb/s

Mobility Makes Things Worse

High rate Video stalls when SNR dipsLow rate Overall video quality is low

Successive frames may experience a different channel

200ms

Time [ms]Rece

ived

Sig

nal L

evel

[dBm

]Mobility causes fast unpredictable SNR variations

Common Problem

Hard to pick a single rate that matches the channel

Wrong bitrate video degrades drastically

But …

In principle, video quality should degrade smoothly with channel quality

Sender should be able to simply transmit:Noisy channel decoded pixels approximate

original pixelsGood channel decoded pixels match originals

Why Cannot Current Design Provide Smooth Degradation?

• Compression and error protection convert real-valued pixels to bits

• Bits destroy the numerical properties of original pixels11110 and 11111 could refer to pixels as different as 5 and 149

• If all bit errors can be corrected all pixels are correct• Even one residual bit error arbitrary errors in pixels

Analog TV Degraded Smoothly

Real-Valued Pixels2, 153, …

Transmitted Values2α, 153α, …

Transmitted values are linearly related to pixel luminance

But Analog TV was not efficient:• No compression• No error protection

α

Small perturbation on channel

Small perturbation in pixel values

It did not convert pixels to bits

SoftCast Combines the Best of Both Worlds

Like Digital TV,It codes for compression and error protection

Like Analog TV,It provides smooth degradation

Goal: transmitted signal is linearly related to the pixels smooth degradation

SoftCast uses a new coding technique that:

– converts pixels to real-valued codewords, not bits

– provides compression and error protection while preserving linearity between pixels and codewords

– passes the codewords to the PHY, which transmits them directly on the channel

SoftCast

Pixels in an image change gradually In frequency domain, most high frequencies are zero

STEP1: Convert a frame to frequency domain using DCT

STEP2: Send only non-zero frequencies in the frame Compressing the frame

How Does SoftCast Compress?

Zeros

DCT ofwhole frame

Encoder needs to tell the decoder the location of zeros– Easy because zeros are clustered

Divide into chunks and drop zero chunks– Use a bit map to tell receiver locations of zero chunks

Drop Zero Chunks

• DCT is a linear operator• Dropping zero chunks does not break linearity SoftCast’s compression preserves linearity

How Does SoftCast Provide Error Protection?

2.5

SoftCast protects real-valued codewords using magnitude-scaling

Codeword Transmitted Received Decoded

24.9

25.1±0.1

2.492.51

±0.01

Channel Noise±0.1

25

x10

Before Tx Scale up

/10

After Rx Scale down

How Does SoftCast Provide Error Protection?

2.5

SoftCast protects real-valued codewords using magnitude-scaling

Codeword Transmitted Received Decoded

24.9

25.1±0.1

2.492.51

±0.01

Channel Noise±0.1

25

x10

Before Tx Scale up

/10

After Rx Scale down

Scaling the codeword up, scales down the effective noise on the channel by the same factor

But Can’t Scale All Codewords UpScaled-up values are larger take more power to transmitBut hardware has limited powerWe find the optimal scaling factors that minimize video errors given hardware power

Theorem • Let λi be the variance of chunk i• The linear encoder that minimizes video errors scales

the values xi in chunk i as follows:

yi = gi xi where gi ~ λi-1/4

Scaling is linear SoftCast’s error protection preserves linearity

How Does the PHY Transmit?

Traditional PHY maps bits to reals (I and Q) using modulation

SoftCast PHY directly transmits the real-valued codewords as I and Q

Recall: Channel transmits pairs of real values (I and Q)

QAM modulation IQ

…0011001

…y[5]y[4]y[3]y[2]y[1]

I

Q

SoftCast achieves its goal of ensuring that the transmitted signal is linearly related to the pixels

…y[5]y[4]y[3]

y[1]

y[2]…y[5]

y[3]y[1]

y[4]y[2]

Performance

Compared Schemes

• SoftCast• MPEG-4 (H.264) over 802.11– Implemented in libx264 via ffmpeg

• 2-Layer Video– A base layer and an enhancement layer– Implemented in libx264 via ffmpeg

Test Setup

WARP

Locations of trace collection

• Collected channel traces with WARP between node in testbed

Test Setup• Collected channel traces with WARP between node in testbed

• Extracted noise patterns as differences between transmitted and received soft values

Trace-Driven Channel

(802.11 OFDM)

MPEG4

2-Layer Video

SoftCast

MPEG4

2-Layer Video

SoftCast

Encoders Decoders

• Compare schemes for the same trace-driven channels

Video Quality vs. Channel Quality

0 5 10 15 20 2520

25

30

35

40

45

Channel Quality – SNR [dB]

Vide

o Q

ualit

y –

PSN

R [d

B]

Video Quality vs. Channel Quality

0 5 10 15 20 2520

25

30

35

40

45

MPEG 6Mbps

Channel Quality – SNR [dB]

Vide

o Q

ualit

y –

PSN

R [d

B]

Video Quality vs. Channel Quality

0 5 10 15 20 2520

25

30

35

40

45

MPEG 6MbpsMPEG 12Mbps

Channel Quality – SNR [dB]

Vide

o Q

ualit

y –

PSN

R [d

B]

Video Quality vs. Channel Quality

0 5 10 15 20 2520

25

30

35

40

45

MPEG 6MbpsMPEG 12MbpsMPEG 18MbpsMPEG 24MbpsMPEG 36MbpsMPEG 48MbpsMPEG 54Mbps

Channel Quality – SNR [dB]

Vide

o Q

ualit

y –

PSN

R [d

B]

MPEG degrades drastically when the bitrate does not match channel SNR

0 5 10 15 20 2520

25

30

35

40

45

SoftCastMPEG 6MbpsMPEG 12MbpsMPEG 18MbpsMPEG 24MbpsMPEG 36MbpsMPEG 48MbpsMPEG 54Mbps

Channel Quality – SNR [dB]

Vide

o Q

ualit

y –

PSN

R [d

B]

SoftCast combines efficiency with smooth video degradation

Video Quality vs. Channel Quality

Multicast

• Receiver 1 has SNR = 5dB – best bitrate 6Mb/s• Receiver 2 has SNR = 21dB – best bitrate 48Mb/s

Multicast

MPEG SoftCast20

25

30

35

40

Vide

o PS

NR

[dB]

• Receiver 1 has SNR = 5dB – best bitrate 6Mb/s• Receiver 2 has SNR = 21dB – best bitrate 48Mb/s

Multicast

MPEG SoftCast20

25

30

35

40

Vide

o PS

NR

[dB]

Layered video:• Base layer at 6Mb/s, enhancement layer at 48 Mb/s• Have to divide medium time between the layers

Multicast

MPEG SoftCast Layered 4:1 Layered 3:2 Layered 2:320

25

30

35

40

Vide

o PS

NR

[dB]

Layered video:• Base layer at 6Mb/s, enhancement layer at 48 Mb/s• Have to divide medium time between the layers

Multicast

MPEG SoftCast Layered 4:1 Layered 3:2 Layered 2:320

25

30

35

40

Vide

o PS

NR

[dB]

In 2-layer video, enhancement reduces transmission time of base Weak receiver becomes worse off

Layered video:• Base layer at 6Mb/s, enhancement layer at 48 Mb/s• Have to divide medium time between the layers

Preliminary Mobility Results

1517192123252729

SNR [dB]

PSN

R [d

B]

7 6.5 6

Preliminary Mobility Results

1517192123252729

MPEG

SNR [dB]

PSN

R [d

B]

7 6.5 6SNR variations cause major glitches in MPEG

1517192123252729

SoftCastMPEG

SNR [dB]

PSN

R [d

B]

7 6.5 6

Preliminary Mobility Results

SoftCast reacts smoothly to changes in SNR

Conclusion

• Digital video can achieve smooth degradation• Key Idea: – Continue to compress and protect against errors– But make codewords linearly related to pixels

• Experimental results show this approach is highly promising for multicast and mobile scenarios