I & Q sample - number of bits

As I understand, I & Q values used in digital communication are not
directly data but they only represent data indirectly. In many systems, I &
Q samples have different sizes (number of bits) and is it generally assumed
that more the bits, better the system.

The question is whether having more bits helps in making better decision in
the receivers as to which of the constellation point the received sample
belongs to?

thanks a lot ...

PS: I am not a DSP or communication expert but these are relevant to me
from an application point ...

	 

_____________________________		
Posted through www.DSPRelated.com

On 12/15/13 9:46 PM, Sharan123 wrote:
> As I understand, I&Q values used in digital communication are not
> directly data but they only represent data indirectly. In many systems, I&Q
> samples have different sizes (number of bits) and is it generally assumed
> that more the bits, better the system.

i think you get better S/N ratio with more bits.  more the S/N, fewer 
the symbol errors.

but i didn't know that there are "many" systems where I and Q samples 
have different word width.

>
> The question is whether having more bits helps in making better decision in
> the receivers as to which of the constellation point the received sample
> belongs to?

sure.  more bits, less roundoff or quantization error.  less error, 
fewer mistaken decisions.


-- 

r b-j                  rbj@audioimagination.com

"Imagination is more important than knowledge."

On 12/15/2013 8:46 PM, Sharan123 wrote:

> The question is whether having more bits helps in making better decision in
> the receivers as to which of the constellation point the received sample
> belongs to?

Good question.

AFAIR it was shown by Viterbi that it is sufficient to have 8 ADC LSB 
steps between the nearest ponts of constellation. This, of course, 
depends on particular system, however 8 LSB separation is ballpark of 
the point of diminishing returns. Very little could be gained by making 
ADC resolution higher then that.

Vladimir Vassilevsky
DSP and Mixed Signal Designs
www.abvolt.com

Thanks a lot ...

Hello R-B-J,

>but i didn't know that there are "many" systems where I and Q samples

> have different word width

I have seen newer basestations where I & Q have word width starting from 11
bits and go all the way upto 24 bits. This is the resolution at basestation
and radio interface.	 

_____________________________		
Posted through www.DSPRelated.com

On Mon, 16 Dec 2013 01:59:19 -0600, "Sharan123" <99077@dsprelated>
wrote:

>Thanks a lot ...
>
>Hello R-B-J,
>
>>but i didn't know that there are "many" systems where I and Q samples
>
>> have different word width
>
>I have seen newer basestations where I & Q have word width starting from 11
>bits and go all the way upto 24 bits. This is the resolution at basestation
>and radio interface.	 

In the cellular basestation case the ADC may need a lot of dynamic
range to handle multiple channels with near and far signals at the
same time.  How many bits are needed depends a lot on what part of the
signal processing chain one is interested in and what sort of system
it is.   As you point out, in some cases a lot of bits are needed and
one can reduce them as you progress through the processing chain, but
in others one might start out with much fewer and increase the number
of bits through processing gain.

I'm not quite sure exactly what the OP was asking about.   If the
question was really just about how many bits are needed at the slicer,
that is a little easier to pin down, but still has a lot of variables;
modulation type, FEC type, etc., etc.


Eric Jacobsen
Anchor Hill Communications
http://www.anchorhill.com

On Sun, 15 Dec 2013 20:46:35 -0600, Sharan123 wrote:

> As I understand, I & Q values used in digital communication are not
> directly data but they only represent data indirectly. In many systems,
> I &
> Q samples have different sizes (number of bits)

Your comment is unclear.  Do you mean that the I channel has a different 
width than the Q channel, or do you mean that system A may have 12 bits 
on I & Q, while system B only has 4?

> and is it generally
> assumed that more the bits, better the system.

In the absence of such piddly little matters like cost and technical 
feasibility, yes.

> The question is whether having more bits helps in making better decision
> in the receivers as to which of the constellation point the received
> sample belongs to?

Yes it does, but you hit the law of diminishing returns.  A pretty good 
model for thinking of this is that when you quantize a signal you add 
noise (so-called "quantization noise").  This quantization noise is added 
on top of whatever noise already exists in the system, and the effective 
total noise amplitude after the addition is the root of the sum of the 
squares (RSS value) of the original noise and the quantization noise 
variances.

Because of the RSS character of the output noise variance, you find that 
once the quantization noise is lower than the already-existing noise in 
the system, its contribution to the output noise drops off rapidly as the 
quantization noise is reduced through adding more bits.

-- 

Tim Wescott
Wescott Design Services
http://www.wescottdesign.com

>In the cellular basestation case the ADC may need a lot of dynamic
>range to handle multiple channels with near and far signals at the
>same time.  How many bits are needed depends a lot on what part of the
>signal processing chain one is interested in and what sort of system
>it is.   As you point out, in some cases a lot of bits are needed and
>one can reduce them as you progress through the processing chain, but
>in others one might start out with much fewer and increase the number
>of bits through processing gain.
>
>I'm not quite sure exactly what the OP was asking about.   If the
>question was really just about how many bits are needed at the slicer,
>that is a little easier to pin down, but still has a lot of variables;
>modulation type, FEC type, etc., etc.

yes, dynamic range was what was mentioned the reason for having higher
number of bits. I should make one correction though in my comment above.

I have seen that the standard itself supporting 11-24 bits for I & Q though
I have not seen systems actually using 24 bits in implementation.	 

_____________________________		
Posted through www.DSPRelated.com

>On Sun, 15 Dec 2013 20:46:35 -0600, Sharan123 wrote:
>I have a question. How can more no of bits correspond to better snr.
Because if a symbol is packed with more no of bits, then bits/symbol is
high. So I need to use a higher order modulation scheme which means my
probability of error increases and hence the distance between constellation
points reduces. I think I am missing something. Maybe I am seeing the no of
bits at someother part of the communication chain. can anybody clarify?
>> As I understand, I & Q values used in digital communication are not
>> directly data but they only represent data indirectly. In many systems,
>> I &
>> Q samples have different sizes (number of bits)
>
>Your comment is unclear.  Do you mean that the I channel has a different 
>width than the Q channel, or do you mean that system A may have 12 bits 
>on I & Q, while system B only has 4?
>
>> and is it generally
>> assumed that more the bits, better the system.
>
>In the absence of such piddly little matters like cost and technical 
>feasibility, yes.
>
>> The question is whether having more bits helps in making better
decision
>> in the receivers as to which of the constellation point the received
>> sample belongs to?
>
>Yes it does, but you hit the law of diminishing returns.  A pretty good 
>model for thinking of this is that when you quantize a signal you add 
>noise (so-called "quantization noise").  This quantization noise is added

>on top of whatever noise already exists in the system, and the effective 
>total noise amplitude after the addition is the root of the sum of the 
>squares (RSS value) of the original noise and the quantization noise 
>variances.
>
>Because of the RSS character of the output noise variance, you find that 
>once the quantization noise is lower than the already-existing noise in 
>the system, its contribution to the output noise drops off rapidly as the

>quantization noise is reduced through adding more bits.
>
>-- 
>
>Tim Wescott
>Wescott Design Services
>http://www.wescottdesign.com
>
>	 

_____________________________		
Posted through www.DSPRelated.com

It depends on what the I and Q samples are used for. If you have a digital equalizer, then yes, the more bits the better. If you do phase tracking in digital domain, also yes. 

If you have a aoft-input ECC after the symbol detector, then yes, you need more than one bit to get useful soft information. 

If you simply use the I and Q samples just to decode a simple modulation (say, QPSK or BPSK) without any digital correction, then perhaps there is no value. 

If the above sounds alien to you, you need to describe your system better so that experts here can walk you step-by-step.


On Sunday, December 15, 2013 9:46:35 PM UTC-5, Sharan123 wrote:
> As I understand, I & Q values used in digital communication are not
> directly data but they only represent data indirectly. In many systems, I &
> Q samples have different sizes (number of bits) and is it generally assumed
> that more the bits, better the system.
> 
> The question is whether having more bits helps in making better decision in
> the receivers as to which of the constellation point the received sample
> belongs to?
> 
> 
> 
> thanks a lot ...
> 
> 
> 
> PS: I am not a DSP or communication expert but these are relevant to me
> 
> from an application point ...
> 
> 
> 
> 	 
> 
> 
> 
> _____________________________		
> 
> Posted through www.DSPRelated.com

Hi,

as was mentioned, you need the bits mainly for adjacent channels and
blockers. For example, any old GSM phone is able to cope with a neighboring
channel at 400 kHz offset close to 50 dB above the own signal 

In general, it's a trade-off with the analog filters before the ADC. 	 

_____________________________		
Posted through www.DSPRelated.com