On 1 Nov 2005 05:38:48 -0800, lindah74uk@yahoo.co.uk wrote:
>Hi Eric,
>
>Thank you for some of the opinions about the OFDM trade-offs study.
>Well, the bill will be too costly to a poor student :-)
>
>Regards,
>Lindah
:) No problem on the bill. I'll just start an account for you... ;)
Just kidding, you're most welcome to the advice.
Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org
Reply by ●November 1, 20052005-11-01
Hi Eric,
Thank you for some of the opinions about the OFDM trade-offs study.
Well, the bill will be too costly to a poor student :-)
Regards,
Lindah
> Lindah,
>
> To me it comes down to equalizer complexity. For OFDM systems the
> equalizer is easily bounded, for a QAM system, especially at a very
> high symbol rate, the equalizer complexity can quickly get out of hand
> if the channel delay spread is very long at all.
>
> It's not difficult to make a first-order tradeoff analysis. For the
> OFDM system add a cyclic prefix to the symbols that's _at minimum_ the
> length of the longest expected channel delay spread, and see how much
> it'll hurt to maybe use 1.5x or 2x that length to accomodate filter
> effects and other stuff that gets in the way. Tradeoff subcarrier
> spacing against the coherence bandwidth and the symbol rates that you
> can process (keeping the CP to 1/3 or less, I like a lot less, of the
> symbol is good for efficiency) and see what the system starts to look
> like. This gives an idea of how big/fast/ugly the FFT, channel
> estimator, etc., are going to be.
>
> Figure out the data rates that you can expect with such a system, and
> then see what it takes to do the same with a single-carrier.
> Whatever the symbol rate turns out to be, the EQ then needs to be
> around the number of taps as T goes into the delay spread. It gets
> ugly in a hurry for high-symbol-rate systems. This is why so many
> wideband systems these days are OFDM if the channel delay spread is
> very long.
>
> You've probably already done some or all of this, so ignore me as
> suitable. But to me that's pretty much the basis of the real
> tradeoffs between OFDM and single carrier. OFDM does, however, let
> you do all kinds of other cool things like adaptive bit loading, power
> loading, OFDMA, etc., that just isn't practical with single carrier
> systems.
>
> Hope that helps a bit or two. Where should I send the bill? ;)
>
>
> Eric Jacobsen
> Minister of Algorithms, Intel Corp.
> My opinions may not be Intel's opinions.
> http://www.ericjacobsen.org
Reply by Snowball●October 28, 20052005-10-28
"Eric Jacobsen" <eric.jacobsen@ieee.org> wrote in message
news:esu4m1pv81dunalatgulhoms0spjf17h5n@4ax.com...
>
> The symbol rate is typically equal to the 3dB bandwidth of the signal
> for QAM. So in 1MHz bandwidth you can have a 1MHz symbol rate, but
> you need some extra on either side for the skirts, and how much more
> is needed depends on the type of filtering that is being employed.
>
> As Mark mentioned, it's not too hard to get a 5.36MHz symbol rate in a
> 6Mhz channel with the rest of it used as guardband for the skirts and
> margin for adjacent channel interference.
>
> You have to do exactly the same thing for OFDM. You can't just pile
> the subcarriers up to the edges of the band, because spectral regrowth
> from the amplifier (and other impairments) still mean that you need
> some guard band, and it winds up being about the same as for a single
> carrier system.
>
> Not only that, but in OFDM you have to slow the symbol rate further
> than you might think in order to accomodate the cyclic prefix or
> cyclic extension. The CP/CE provide lots of goodness for multipath
> handling, but are dead weight as far as information transmission is
> concerned. For this reason sometimes when somebody waves the "OFDM is
> more spectrally efficient" card, I wave the "no it's not, it's LESS
> spectrally efficient" card just to make the comparison. In most
> cases when it gets down to the real engineering there is little or no
> difference in spectral efficiency between OFDM and single carrier
> systems in my experience.
>
> Eric Jacobsen
> Minister of Algorithms, Intel Corp.
> My opinions may not be Intel's opinions.
> http://www.ericjacobsen.org
Hi Eric,
Thanks for the answer, I realized I was wrong after I had a look at the
pre-adsl modem protocols (e.g. v.34)... Anyway, I agree with you 100%.
Reply by Jerry Avins●October 28, 20052005-10-28
Mark wrote:
...
> Jerry,
>
> I am referring to digital 64QAM CATV signals, not analog CATV signals
> which are VSB as you state.
I thought so at first, but your reference to 6 MHz channels threw me
off. Old fogy and all that.
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by Mark●October 28, 20052005-10-28
Jerry Avins wrote:
> Mark wrote:
> > Jerry Avins wrote:
> >
> >>Mark wrote:
> >>
> >>>Snowball wrote:
> >>>
> >>>
> >>>>"Mark" <makolber@yahoo.com> wrote in message news:1130430253.588590.265420@g43g2000cwa.googlegroups.com...
> >>>>
> >>>>
> >>>>>I don't see your point about OFDM ??
> >>>>>
> >>>>>If you have 1 million carriers each 1 Hz wide, then each carrier can
> >>>>>support a symbol rate of 1/2 sym/s. The total aggregate symbol rate is
> >>>>>still 500ksym/s just like QAM?
> >>>>>
> >>>>>Mark
> >>>>>
> >>>>
> >>>>Hi Mark,
> >>>>
> >>>>If the symbol rate, 1/T, on each of the subcariers in OFDM is equal to the subcarrier separation, they are orthogonal over the symbol interval T. Thus if you were to place 1 million carriers in 1MHz, they would be spaced at 1Hz and their symbol rate would be 1symbol/s. I'm not sure where you got 1/2 symbols/s.
> >>>
> >>>
> >>>OK you are correct, my bad... I asked the wrong question...
> >>>
> >>>For the QAM case, why are you saying that QAM can only support
> >>>500ksym/sec in a 1 MHz bandwidth. Cable TV achieves 5.36 Msps in a 6
> >>>MHz BW?
> >>
> >>Single sideband.
> >>
> >>Jerry
> >>--
> >>Engineering is the art of making what you want from things you can get.
> >
> >
> >
> > Nope, QAM on CATV is not single sideband, the carrier is in the center
> > of the channel.
>
> An analog cable signal works without modification right into a standard
> NTSC receiver. While not true SSB, vestigial sideband comes close.
>
> Jerry
Jerry,
I am referring to digital 64QAM CATV signals, not analog CATV signals
which are VSB as you state.
Mark
Reply by Eric Jacobsen●October 28, 20052005-10-28
On 27 Oct 2005 04:53:25 -0700, lindah74uk@yahoo.co.uk wrote:
>Hi Eric,
>
>The channel is freqeuncy selective as shown by many measurements. As
>pointed out by Allan that we have a huge bandwidth, then a less
>efficeint could be used. So the question is which modulation will give
>the best trade-off among the the constraints mentioned above.
>
>I guess it might not be so easy to answer unless simulation is
>performed in a realistic channel and circuit models. But I would be
>interested to hear some opinions and discussions from the experts in
>this newgroup about the suitability of type of modulation subject to
>the aforementioned problems.
>
>Comments are very much welcome.
>
>Regards,
>Lindah
Lindah,
To me it comes down to equalizer complexity. For OFDM systems the
equalizer is easily bounded, for a QAM system, especially at a very
high symbol rate, the equalizer complexity can quickly get out of hand
if the channel delay spread is very long at all.
It's not difficult to make a first-order tradeoff analysis. For the
OFDM system add a cyclic prefix to the symbols that's _at minimum_ the
length of the longest expected channel delay spread, and see how much
it'll hurt to maybe use 1.5x or 2x that length to accomodate filter
effects and other stuff that gets in the way. Tradeoff subcarrier
spacing against the coherence bandwidth and the symbol rates that you
can process (keeping the CP to 1/3 or less, I like a lot less, of the
symbol is good for efficiency) and see what the system starts to look
like. This gives an idea of how big/fast/ugly the FFT, channel
estimator, etc., are going to be.
Figure out the data rates that you can expect with such a system, and
then see what it takes to do the same with a single-carrier.
Whatever the symbol rate turns out to be, the EQ then needs to be
around the number of taps as T goes into the delay spread. It gets
ugly in a hurry for high-symbol-rate systems. This is why so many
wideband systems these days are OFDM if the channel delay spread is
very long.
You've probably already done some or all of this, so ignore me as
suitable. But to me that's pretty much the basis of the real
tradeoffs between OFDM and single carrier. OFDM does, however, let
you do all kinds of other cool things like adaptive bit loading, power
loading, OFDMA, etc., that just isn't practical with single carrier
systems.
Hope that helps a bit or two. Where should I send the bill? ;)
Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org
Reply by Eric Jacobsen●October 28, 20052005-10-28
On Thu, 27 Oct 2005 13:43:54 +0200, "Snowball" <sdris@softlab.ntua.gr>
wrote:
>"Eric Jacobsen" <eric.jacobsen@ieee.org> wrote in message
>news:ogcvl1lvfbfviikp7vkfkqnrbnm9l3uo64@4ax.com...
>> On 25 Oct 2005 23:54:06 -0700, lindah74uk@yahoo.co.uk wrote:
>>
>> I often see OFDM touted as "spectrally efficient", but it's only
>> efficient if the channel is frequency selective. If the channel is
>> AWGN with a reasonable SNR then a high-order single carrier QAM system
>> may be better.
>
>Correct me if I am wrong, but isn't it the case that the symbol rate for QAM
>can be only as high as the carrier frequency? This limits the channel
>bandwidth you can cover, and hence the total data-rate you can achieve. For
>example, assume you have a flat fading channel that can pass frequencies
>from DC to 1MHz. If you place the QAM carrier at 500kHz, your symbol rate
>would have to be 500Ksymbols/s max. With OFDM you can fill up (almost) the
>entire 1MHz with carriers, hence reaching an aggregate symbol rate close to
>1Msymbol/s. Is this correct...?
The symbol rate is typically equal to the 3dB bandwidth of the signal
for QAM. So in 1MHz bandwidth you can have a 1MHz symbol rate, but
you need some extra on either side for the skirts, and how much more
is needed depends on the type of filtering that is being employed.
As Mark mentioned, it's not too hard to get a 5.36MHz symbol rate in a
6Mhz channel with the rest of it used as guardband for the skirts and
margin for adjacent channel interference.
You have to do exactly the same thing for OFDM. You can't just pile
the subcarriers up to the edges of the band, because spectral regrowth
from the amplifier (and other impairments) still mean that you need
some guard band, and it winds up being about the same as for a single
carrier system.
Not only that, but in OFDM you have to slow the symbol rate further
than you might think in order to accomodate the cyclic prefix or
cyclic extension. The CP/CE provide lots of goodness for multipath
handling, but are dead weight as far as information transmission is
concerned. For this reason sometimes when somebody waves the "OFDM is
more spectrally efficient" card, I wave the "no it's not, it's LESS
spectrally efficient" card just to make the comparison. In most
cases when it gets down to the real engineering there is little or no
difference in spectral efficiency between OFDM and single carrier
systems in my experience.
Eric Jacobsen
Minister of Algorithms, Intel Corp.
My opinions may not be Intel's opinions.
http://www.ericjacobsen.org
Reply by glen herrmannsfeldt●October 28, 20052005-10-28
Jerry Avins wrote:
> Mark wrote:
>(snip)
>> OK you are correct, my bad... I asked the wrong question...
>> For the QAM case, why are you saying that QAM can only support
>> 500ksym/sec in a 1 MHz bandwidth. Cable TV achieves 5.36 Msps in a 6
>> MHz BW?
> Single sideband.
I haven't looked at this recently, but I thought it could change at
twice the carrier frequency, and synchronous to the carrier.
I think, then, that would result in double sideband with different
signals on each sideband.
Having not done the calculation recently I am not so sure, though.
As I understand it a sufficient number of transitions are required for
the decoder to remain synchronized, so that scramblers are usually used
on the encoder.
-- glen
Reply by Jerry Avins●October 28, 20052005-10-28
Mark wrote:
> Jerry Avins wrote:
>
>>Mark wrote:
>>
>>>Snowball wrote:
>>>
>>>
>>>>"Mark" <makolber@yahoo.com> wrote in message news:1130430253.588590.265420@g43g2000cwa.googlegroups.com...
>>>>
>>>>
>>>>>I don't see your point about OFDM ??
>>>>>
>>>>>If you have 1 million carriers each 1 Hz wide, then each carrier can
>>>>>support a symbol rate of 1/2 sym/s. The total aggregate symbol rate is
>>>>>still 500ksym/s just like QAM?
>>>>>
>>>>>Mark
>>>>>
>>>>
>>>>Hi Mark,
>>>>
>>>>If the symbol rate, 1/T, on each of the subcariers in OFDM is equal to the subcarrier separation, they are orthogonal over the symbol interval T. Thus if you were to place 1 million carriers in 1MHz, they would be spaced at 1Hz and their symbol rate would be 1symbol/s. I'm not sure where you got 1/2 symbols/s.
>>>
>>>
>>>OK you are correct, my bad... I asked the wrong question...
>>>
>>>For the QAM case, why are you saying that QAM can only support
>>>500ksym/sec in a 1 MHz bandwidth. Cable TV achieves 5.36 Msps in a 6
>>>MHz BW?
>>
>>Single sideband.
>>
>>Jerry
>>--
>>Engineering is the art of making what you want from things you can get.
>
>
>
> Nope, QAM on CATV is not single sideband, the carrier is in the center
> of the channel.
An analog cable signal works without modification right into a standard
NTSC receiver. While not true SSB, vestigial sideband comes close.
Jerry
--
Engineering is the art of making what you want from things you can get.
�����������������������������������������������������������������������
Reply by Mark●October 27, 20052005-10-27
Jerry Avins wrote:
> Mark wrote:
> > Snowball wrote:
> >
> >>"Mark" <makolber@yahoo.com> wrote in message news:1130430253.588590.265=
420@g43g2000cwa.googlegroups.com...
> >>
> >>>I don't see your point about OFDM ??
> >>>
> >>>If you have 1 million carriers each 1 Hz wide, then each carrier can
> >>>support a symbol rate of 1/2 sym/s. The total aggregate symbol rate is
> >>>still 500ksym/s just like QAM?
> >>>
> >>>Mark
> >>>
> >>
> >>Hi Mark,
> >>
> >>If the symbol rate, 1/T, on each of the subcariers in OFDM is equal to =
the subcarrier separation, they are orthogonal over the symbol interval T. =
Thus if you were to place 1 million carriers in 1MHz, they would be spaced =
at 1Hz and their symbol rate would be 1symbol/s. I'm not sure where you got=
1/2 symbols/s.
> >
> >
> > OK you are correct, my bad... I asked the wrong question...
> >
> > For the QAM case, why are you saying that QAM can only support
> > 500ksym/sec in a 1 MHz bandwidth. Cable TV achieves 5.36 Msps in a 6
> > MHz BW?
>
> Single sideband.
>
> Jerry
> --
> Engineering is the art of making what you want from things you can get.
Nope, QAM on CATV is not single sideband, the carrier is in the center
of the channel.
Mark
Mark