Application Note

AN-1177

Rev. 0 | Page 9 of 12

part-to-part skew is the greater of these differences (in the case

INPUT

ACTUAL

OUTPUT

ACTUAL

OUTPUT

(2ND)

ACTUAL

OUTPUT

(3RD)

ACTUAL

OUTPUT

(4TH)

–

–

CHANNEL-TO-CHANNEL

OR PART-TO-PART SKEW

(

>

D–

D+

D–

D+

D–

D+

D–

D+

D–

D+

Figure 18. Waveforms Illustrating Channel-to-Channel or Part-to-Part Skew

Both channel-to-channel skew and part-to-part skew result in

parallel data channels received out of phase relative to each

other, even if they were synchronized at the transmitting end.

This can cause difficulties in sampling across multiple channels.

DATA ENCODING AND SYNCHRONIZATION

The challenges for LVDS timing stem not only from the

high speed transmission, but also from the data encoding.

In many LVDS applications, in order to increase bandwidth,

multiple parallel LVDS channels are used to transmit data.

The transmitter must synchronize data transmitted on these

channels and the receiver needs to sample each channel at the

appropriate point so that data can be received at the same time

across channels.

In LVDS applications using few channels, serial data is typically

transmitted and at higher speeds. The high speed requires the

receiving device to synchronize quickly with the incoming data

stream, and, in addition to accurately sampling each bit, the

receiving device needs to detect frames of data in the incoming

bit stream.

To help the receiving device synchronize with the received data,

a clock may be transmitted with the data channels. This is

described as source-synchronous data transmission. There are

several methods of transmitting the clock with the data. The

clock may be transmitted as a parallel channel, with the clock

period corresponding to one data bit (single data rate, SDR)

or two data bits (double data rate, DDR). For serial LVDS

transmission, a frame clock may also be transmitted. An

example of ADC source-synchronous LVDS outputs for SDR

and DDR is shown in Figure 19.

SAMPLE N

SAMPLE N + 2

SAMPLE N + 1

CLK+

SAMPLE N – 7

BIT 0 (LSB)

BIT 0

(LSB)

BIT 5

BIT 0

(LSB)

BIT 5

BIT 9

(MSB)

BIT 4

BIT 9

(MSB)

BIT 4

SAMPLE N – 7

SAMPLE N – 6

BIT 0 (LSB)

CLK–

DCO+

DCO–

D0+

D0–

SAMPLE N – 7

BIT 9 (MSB)

SAMPLE N – 6

BIT 9 (MSB)

D9+

D9–

ANALOG

INPUT

INTERNAL CLOCK:

LVDS OUTPUTS:

SDR

(10 CHs)

DDR

(5 CHs)

D0/D5+

D0/D5–

D4/D9+

D4/D9–

SAMPLE N – 6

Figure 19. ADC input and Source-Synchronous LVDS Output Waveforms

An alternative to dedicated clock channels is to embed the clock

with the data. With the embedded clock method, fixed bits are

inserted into the data stream, allowing a receiving node to

detect these bits and synchronize with the incoming data.

Channel-to-channel and part-to-part skew can be compensated

for when received by modern FPGAs, using a scheme termed

dynamic phase adjustment (DPA). The FPGA generates

multiple phases of the received source-synchronous clock and

matches each data channel to the best clock phase for sampling.

If DPA is not available, then a strict timing budget must be

adhered to. There must be a time interval remaining after

transmitter channel-to-channel skew and the sampling time

are subtracted from the bit period. This interval is termed the

receiver skew margin. The transmitter channel-to-channel skew

includes the skew across channels due to the transmitting node,

the skew due to the medium and the clock skew relative to

the data.