how to compare the two phase difference width for two signals?

[xworld2008]

as below:

signal a______________|-----------------------|__________________
signal b___________________|-------------------------|___________
- Ta - - Tb -

Ta,Tb are less than 100p, Ta, Tb are changing when others signals are changing, How to compare Ta and Tb(Ta > Tb ?)?
no higher clock is available.

Thanks.

 

[balucyber124]

may i know what 100p exactly is?

 

[jeffrey samuel]

100p is 100 pico seconds pal

 

[xworld2008]

Yes. now we use two buffer chains which one is smaller, one is bigger. then connect a to small one. b to bigger one. then connect buffer output to FF. then compare the value to two FF groups. this method will take big area. also the setup/hold time will effect the accuracy.
any one have better way?

 

[xworld2008]

anyone have idea?

 

[xworld2008]

Can someone help me?

 

[nba83]

Yes. now we use two buffer chains which one is smaller, one is bigger. then connect a to small one. b to bigger one. then connect buffer output to FF. then compare the value to two FF groups. this method will take big area. also the setup/hold time will effect the accuracy.
any one have better way?

for my information, how you are simulating your design?

 

[qieda]

Hi,

maybe you can do the phase detect with something like in the picture.
You use one input as a clock and the second input as a data input.
You need a unit delay. The clock input is delayed by 2 unit delays.
The data is delayed by 0,1,2,3,4 unit delays. For stability I would latch the output of the 5 input samplers with a second stage (maybe the falling edge of the clock input, assuming your wavetrace is correct)
After that you can use a decoding logic to find the phase



best regards

 

[xworld2008]

the difference value is less than one unit delay, how can i get the comparison result using this phase detect?

 

[qieda]

Hi,

If your unit delay is larger than the difference, you can not use this circuit.

whats your unit delay? In current ASIC technologies I would assume you can use an inverter which should be around 50ps.

A different approach could be to use every input as a clock and sample the other input (see picture)



if you get a b10 or b01 you can decode which input was first.
if you get b00 or b11 your input are nearly equal.
This depends on the setup/hold of the selected flipflops

regards

 

[xworld2008]

Thanks, The difference is less that a unit delay(buffer delay), so this is not good way. currently I just use two different size buffer chains, we can utilize the different between two buffers to capture the difference. But this ciruit is too big, I want to find a simple method to do it.

Best Regards,