peak to average power ratio (PAPR)

(This question is coming from an OFDM perspective but is obviously not
confined to OFDM)

If you are computing the peak to average power ratio (PAPR) of a block
of data, do you include the peak in computing the average power? For
example, do you compare the peak power with the average of all the
rest? Or do you compare the peak with the whole block including itself?

I am pretty sure that it is the latter, judging by some papers I have
read, but I would just like to see if there is a consensus on this here
in comp.dsp.

Slainte
Porterboy

<porterboy76@yahoo.com> wrote in message 
news:1126594755.866583.128220@f14g2000cwb.googlegroups.com...
> (This question is coming from an OFDM perspective but is obviously not
> confined to OFDM)
>
> If you are computing the peak to average power ratio (PAPR) of a block
> of data, do you include the peak in computing the average power? For
> example, do you compare the peak power with the average of all the
> rest? Or do you compare the peak with the whole block including itself?
>
> I am pretty sure that it is the latter, judging by some papers I have
> read, but I would just like to see if there is a consensus on this here
> in comp.dsp.


In cellular, we use the latter. It is relatively easy to have an instrument 
find the long term average and the peaks. It takes more effort, to 
recalculate the average after omitting the peaks, plus a question arises as 
to how much of the peak do you leave out.

So just find the peak and the long term average and then their ratio. This 
is how I've seen it done for AMPS, iDEN, TDMA, GSM, and CDMA protocols.



Clay

porterboy76@yahoo.com wrote:
> example, do you compare the peak power with the average of all the
> rest? Or do you compare the peak with the whole block including itself?

As Clay said, the latter is definitely right. However, I rather wonder 
why the former can be a measure for PAPR. Average power represent the 
average of all signal powers not of some specific powers. Do you have 
any example where the average is calculated except some conditional cases.

-James GOLD

<porterboy76@yahoo.com> wrote in message 
news:1126594755.866583.128220@f14g2000cwb.googlegroups.com...
> (This question is coming from an OFDM perspective but is obviously not
> confined to OFDM)
>
> If you are computing the peak to average power ratio (PAPR) of a block
> of data, do you include the peak in computing the average power? For
> example, do you compare the peak power with the average of all the
> rest? Or do you compare the peak with the whole block including itself?
>
> I am pretty sure that it is the latter, judging by some papers I have
> read, but I would just like to see if there is a consensus on this here
> in comp.dsp.
>
> Slainte
> Porterboy
>

Definitely the latter, however, make sure that your measurement techniques 
are in-line with the reason why you care about peak power in the first 
place. For instance, say I am detecting peak power across a 50 ohm resistor 
by measuring the voltage with an ultra fast ADC (>1GHz full power bandwidth) 
and I have a huge voltage spike in just one of my samples. Should this be 
considered the peak? The spike could be as short as 1ns and not likely cause 
a change in temperature of the resistor, and probably wouldn't even show up 
on FCC test equipment. Hope you see where I'm going with this.

The nice thing about DSP is that you can filter the data and pick the 
appropriate measurement algorithm.

Thomas 

Clay S. Turner wrote:
> <porterboy76@yahoo.com> wrote in message
> news:1126594755.866583.128220@f14g2000cwb.googlegroups.com...
> > (This question is coming from an OFDM perspective but is obviously not
> > confined to OFDM)
> >
> > If you are computing the peak to average power ratio (PAPR) of a block
> > of data, do you include the peak in computing the average power? For
> > example, do you compare the peak power with the average of all the
> > rest? Or do you compare the peak with the whole block including itself?
> >
> > I am pretty sure that it is the latter, judging by some papers I have
> > read, but I would just like to see if there is a consensus on this here
> > in comp.dsp.
>
>
> In cellular, we use the latter. It is relatively easy to have an instrument
> find the long term average and the peaks. It takes more effort, to
> recalculate the average after omitting the peaks, plus a question arises as
> to how much of the peak do you leave out.
>
> So just find the peak and the long term average and then their ratio. This
> is how I've seen it done for AMPS, iDEN, TDMA, GSM, and CDMA protocols.
>
>
>
> Clay

I've seen peak RF power measured with a spectrum analyzer set with span
= RBW and peak hold turned on. The trace will drift upward and little
upticks will appear in it. The highest one is the peak power. I can
remember feeding big signals into a microphone jack and tweaking an
audio compressor to constrain the peak power to FCC limits.

The average power is inclusive of everything.

John

Hi James,

I am using a peak detector (in time) as a synchronisation algorithm. My
definition of a peak in this algorithm is when the instantaneous power
exceeds the rolling average power by a given factor. So either approach
will work for me, provided I define the given "factor" correctly.

The reason I asked the question is that when I eventually have to
explain my algorithm, I would like to use a definition of PAPR that
wont raise eyebrows. It's a small thing really, but I find it better
not to give the cynics a foothold.

Slainte
Porterboy