ISSCC 2011 Short Course Transcription

System Requirements of RF Transceivers

Instructor: Hooman Darabi

1. System Requirements

Good morning everyone, and thank you for attending this short course. As John mentioned, this part of the
short course, the topic is the System Requirements for the RF Receivers and how specifications essentially
define in standards will translate to what the circuit designers need to design the receivers and transmitters.
So it’s mainly a system topic, not much of a circuit for this section.

2. Outline

So, we’re going to talk about some very basic, general requirements of the RF transceivers, mostly well
known material but it’s going to establish the background for what is coming later on.

Then we’re going to talk about a brief overview of the standards, mostly we are going to focus on cellular
wide area network standards, as well as local area where they stand for this presentation, and then we’ll talk
about transceiver architectures and how these numbers translate to specific architectures for specific
applications. And at the same time, we will derive the circuit aspects of RF transceivers, given the system
requirements defined by this standard. And toward the end, we will talk about some more advanced topics
such as next generation transceiver standards and the handset calibration.

3. General Noise and Linearity Concerns

4. Mobile Communication & Cellular Networks

So, we’re going to start with the basic part, essentially a wireless environment such as a cellular as shown
here, consists of several bay stations which communicate through the handsets. Now the area of interest is
divided into several cells, each of them today, probably a mile or so wide, and in order to establish good
communication, and this case, basically a successful phone call, or data reception, there are several
requirements, both from receiver stand point as well as the transmitter. Now for a good reception, we need a
good sensitivity that defines the range, or how far the handset will be from the bay station in a given cell, and
also the resilience of the receiver to other interfering signals which are inevitable in such a wireless network.
Then as far as the transmitter goes, we have the parameter such as the output power which is also the
opposite of sensitivity; again, it defines the range of operation and the transmitter quality as well as the
transmitter interference in noise which will affect the other users.

5. Noise Figure and Sensitivity

So we’re going to talk about some of these aspects in a bit more detail even though most of it is well known
material so the most, the foremost, and most important aspect of a receiver is defined at its sensitivity which
is essentially the ability of the receiver to detect a certain weak signal and yet establish a good receive quality
or certain signal to noise ratio. Now by definition, the noise figure is signal to noise ratio at the input divided
by signal to noise ratio at the output, it’s a log domain calculation, so it can be expanded as given here. Now
the input is a source resistance of Rs, so the noise is the voltage in power domain divided by 4 times the
resistance, the noise of the resistor as we all know is 4 KT times the resistance. So it simplifies to something
like this, the KT is a known quantity, only a function of temperature. The noise is dBc/Hz, so we integrate
over a certain bandwidth, to convert it to dBm or mW.

Now KT in the log domain is -174 at the room temperature, so our equation here, simplifies to this one, and
what it says is that what kind of noise figure is needed for the system to establish a given SNR, or reception
quality for a given signal at the input, which in this case will be essentially our sensitivity.

The bandwidth and the SNR are set by the standards; they are well known quantities, and most of the times,
out of our control. So as RF designers, what we are left with is the noise figure for a given sensitivity which is
derived in the standard and that’s merely a function of what kind of range you’re going to cover.

6. In/Out-of-Band Blockers

Now in addition to sensitivity which defines the range for the desirable signal, there are also other users
nearby which will affect the receiver of interest. Particularly, in the wide area network applications such as
cellular, the blockers are quite stringent and often times quite troublesome. That’s mainly because of the fact
that the handset of interest could happen to be far from the bay station of interest so it’s receiving a weak
desired signal, shown in the green, but it could receive strong signals from other bay stations like the red ones
shown here. Now there’re two types of interferes, or blockers, or jammers; one is from our own standard of
interest, in this case being the other bay stations nearby, or it could be any random wireless signal which is
always present in the environment. This could be something as simple as a microwave oven turning on, or a
wireless band access point, or a GPS signal or anything else for that matter.

Now these blockers essentially define three parameters, three extra parameters besides the noise figure for the
receiver, namely the selectivity or the amount of filtering which is needed, as well as the linearity and phase
noise. And we’re going to talk about these in a bit more detail in the next following slide.

But note that the filtering usually consists of two portions, one is a front-end filter, which is mostly going to
pass through the in-band blockers because the filter cannot be too short to filter them, but it’s going to reject
the outer-band blockers signals belonging to other transmitters, not the standard interest. While the in-band
blockers are coming through and eventually they will be filtered through some back-end filtering or some IF
filtering which will eventually select the desirable signal and reject all of these other blockers.

Now, the blockers affect the receivers in three different ways or there are three sets of parameters which are
defined for a given blocker profiled in a receiver.

7. Reciprocal Mixing

One is the phase noise which arises from the reciprocal mixing. It’s also fairly well known to many of you that
to ease some of the signal processing we have in the receiver, we usually use down converters and the local
oscillator signal or the LO which is used to down convert the desired signal as well as the blocker which is
inevitably sitting there. It is not perfect, there is some noise associated with it, known as phase noise, as
opposed to additive noise here, and this will reciprocally down convert the blockers so it will establish a noise
floor which is integrated in the bandwidth of interest, lying on top of the desirable signal. Again, it’s a fairly
well known equation, which relates the phase noise needed to establish a certain SNR or to establish a certain
quality of reception, given the relative strength of the blocker compared to the desirable signal. Now of
course the phase noise in dBc/Hz so we’ll always integrate over the bandwidth of interest to establish a dBm
number.

Now this equation can be rewritten and recalculated as shown here, so it’s the same thing but it’s defined in
terms of noise figure. If we have a certain blocker present in the receiver, and there’s a certain associated
phase noise, which is coming from the imperfect local oscillator signal, then we can define a noise figure for
the receiver. Now this noise figure is different from what I just showed you earlier. This is merely a function
of the blocker, if there is no blocker present, noise figure is 0dB, the system is perfect, as far as this aspect is
concerned, but because of having the blocker in certain phase noise, the receiver noise floor is elevated. So
we’re going to use this equation to calculate some of the phase noise requirements of our transceiver later on.

8. Gain Compression & Desensitization

Now the second way that the blockers can affect a receiver is through compression. Again, it’s fairly well
known that most of the systems, the receivers that we use, they have an output to input or gain characteristics
that’s shown here, which is compressive, meaning that as the input signal increases, the gain is going to
decrease. In most cases, the desirable signal is weak, so we’re operating here, and of course by means of
some gain control, we can always adjust the signal as such, that we are well below the compression point.
But the problem is that there are scenarios that we explained earlier, that a desirable weak signal is
accompanied by an inevitable large blocker and how this blocker will drive the gain to operate closer to the
saturation point. Now in this case, for example shown here, we have the desirable signal power which is
weak, but also a large blocker, and it’s going to operate here, and therefore effectively reduces the slope or
gain here, and thus, desensitizes the receiver. Meaning that since the front-end gain is smaller, than the noise
figure is going to increase. And there’s a simple equation that can be easily calculated which shows the
desensitized gain as a functional blocker.

So this is another important aspect that must be considered when designing a receiver, even though we might
have perfect phase noise or perhaps perfect 0dB noise figure, simply because of having the blocker and the
effective gain reduction, but we’ll see elevated noise in the receiver.

9. Third Intercept Point

And finally, the third mechanism that blockers can affect us is through intermodulation. The most well
known case is the third order intermodulation, or the third intercept point, IIP3. If we have two blockers
close by, because of the third order nonlinearity of the system, what we’ll have an IIP3 tone, or a third
intermodulation which could fall on the top of the weak desirable signal and it establishes a noise source.
Again, the IIP3 is a fairly known quantity as you all know it will be derived based on this equation and this
kind of approximate equation. But what is more useful is to actually reconvert this to something more
meaningful, as far as a receiver is concerned, which relates to IIP3 for a given blocker to the minimum signal
that can be tolerated to establish a certain SNR.

So this means that if the receiver suffers from third order nonlinearity, even though it might be perfect
otherwise, there will be a noise floor and a sensitivity associated with it which is merely coming from the third
order intercept point problem. Now the intermodulation can be any degree, like I said the third degree is
probably the most common and the most troublesome.

10. Second-Order Distortion in Receivers

Also, we have second-order distortion, mostly similar mechanism. If we have a blocker going through a
system which has second-order nonlinearity, it will create an interfering signal at or around DC. If the
blocker is a static signal, it’s simply a static dynamic DC offset, but if the blocker is amplitude modulated
which is often times the case, it will actually create a spectrum around zero, and that also establishes a noise
floor, so eventually we can relate the IIP2 which is characterized very similar to IIP3 to a certain sensitivity
for a given SNR in the receiver.

Now all these pre mechanisms, they are tightly related to the selectivity of receiver, for the amount of filtering
that we can achieve, so any of these parameters, may be relaxed if we can guarantee a certain amount of
filtering. Mostly because these blockers will be attenuated so the linearity or phase noise or compression
requirements of the receiver will be relaxed. So there’s always a tight tradeoff or tight compromise between
defining the linearity, phase noise and compression point of the receiver as opposed to how much filtering can
be achieved at front-end or IF which is mostly the latter dictated by the cost and part consumption
constraints. So it’s typically a collaborated effort here.

11. Transmitter Mask and Spurs

Now moving to the transmitter, we have more or less the same requirements, but from the opposite point of
view. Meaning that we need to have a certain output power and that is the dual of sensitivity, it essentially
defines the range. The transmitter must establish a certain quality of signals such that the receiver can
successfully receive that, despite the fact that it is also subject to its own noise and blocking scenarios. As
well as certain masks, which is typically associated with the desirable transmitter signal, such that it does not
cause substantial or significant interference for the other users.

So when we’re designing the receiver, we’re always worried or concerned, how much tolerance we have to
other noise sources, while we are designing our transmitter, we are always cautious how much interference
we are causing for the other nearby receivers.
So there’s usually a mask and noise floor which is associated with the transmitter to ensure that the other
nearby receivers are not affected significantly. And again, for a given standard, it’s a collaborated effort
between how good of a transmitter we need for a given receiver and where we have to set the boundaries and
tradeoffs.

12. Phase Noise in Transceivers

Now, for the transmitter, the phase noise, again affects the performance for those two concerns that I
mentioned, essentially the transmitter quality which is defined as an EVM or phase error, mostly set by the
in-band phase noise requirement of our system, whereas the far-out usually sets the mask along with the
linearity of transmitter, meaning that how much interference our transmitter is going to generate to the others.

13. Introduction to Wireless Standards

So with this, fairly quick and simple background, now we’re going to have an eye on the wireless standards.
What is defined in the standards? Why and how it’s going to be translated to more meaningful and tangible
numbers, such as noise, or IIP3 as far as the circuit designer’s concern.

Like I said, I’m going to mostly cover cellular and wireless LAN for this presentation. Mainly because they
are the most challenging one and everything else can be calculated similarly.

14. Evolution of Wireless Standards

Now, like many other applications for the wireless system, the evolution is driven by the need to achieve
higher throughput or data rates. So early in the nineties, we defined GSM, global system for mobile
communication, which was essentially only a voice driven standard, followed by GPRS which was a packet
based standard and it could guarantee some data reception. And throughout many years, it has been evolved
to guarantee more and more throughput, so the follow up to GSM was EDGE, or enhanced data rate for GSM
evolution.

Then we had EDGE evolution which was enhanced further in terms of data rate. And in case you haven’t
noticed, the Y axis here is our mobility or the range of coverage, and our X axis here is the throughput, or the
data rate we can achieve. Now later, in probably late nineties, early 2000s, we introduced 3G standards
which consist of wideband CDMA or UMTS, as well as HSPA, HSDPA, high speed downlink packet access,
which they are using, moving to this direction, they are using more complex modulation schemes, as well as
wider bandwidth to establish higher throughputs, yet maintaining the same mobility.

Now these on top are the cellular standards, now the latest is LTE or long term evolution which can guarantee
as much as a 100 Mbps downlink speed and still good mobility.

Shown in the bottom is the wireless LAN family of standards which achieves similar kind of throughputs but
with much lower mobility or much shorter ranges. We have the wireless family LAN family of a/b/g/n and
the WiMAX which actually is in terms of the modulation and bandwidth, is fairly similar to LTE, or long term
evolution, they both guarantee in the vicinity of 100Mbps, the best case data rate. The main difference
between LTE vs. WiMAX is the mobility, meaning that the LTE can guarantee faster speeds yet, maintaining
good communication whereas the WiMAX and wireless LAN family are mainly local area network with short
term applications.

15. WLAN

So now, starting with wireless LAN again, and it’s fairly well known, we have 2 bands of operation, 2.4G ISM
band as well as the 5GHz which is the 11a vs. b/g. It uses OFDM modulation at 64 QAM, and in case of
11a/g, it can support up to 54Mbps. This is a 20 MHz baseband bandwidth, plus minus 10MHz and is single
input, single output, it was later extended to 11n which is covering wider bandwidth and multiple input,
multiple output which will either guarantee a better reception or a higher throughput depending on the signal
condition.

Shown here is for example, the 11g spectrum which consists of 14 channels span over 80MHz ISM band.

16. Concept of MIMO

Now, what is interesting is the addition of the MIMO, or multiple input, multiple output, so that wireless LAN
standards and then later on, to the cellular in fact. The idea here is that if we have multiple antennas in
transmitters as well multiple antennas in the receivers, we can benefit from two distinct advantages. If the
reception is poor, for example, if you’re subject to fading, then we can have spatial diversity, meaning that we
can receive two different types of signal which are subject to two different types of fading. For example, red
and blue as shown here, and once it’s received and combined, it can actually establish a better quality
signal. So this essentially guarantees our quality when the reception is poor, or on the other extreme, if the
reception is good, because of adding different signals, it can guarantee higher throughput. So it’s trading the
robustness vs. the data rate, depending on signal condition. And that’s most of the wireless LAN systems
today, are benefiting from this feature.

17. WLAN Transceiver Requirements

Now, the requirements for the wireless LAN is mainly driven by the range of coverage and the throughput.
The range of coverage, like we talked about, it translates to the noise figure for the receiver and output
power for the transmitter, and the throughput is mainly a function of the SNR or the reception transmitter
quality. If the signal is strong, meaning that we’re away from sensitivity or noise figure concerns, and then
what matters is the SNR that is established here.

So shown here is for example, a table which summarizes the rate vs. the sensitivity, we have assumed 2dB of
front-end loss and 3dB of noise figure, combined we have 6dB of noise figure, and it shows several scenarios,
wireless LAN systems are adaptive, meaning that if the signal quality of strength is improving, we can also
improve the data rate.

So proportionally, the sensitivity, inversely actually is proportional to the data rate. Also shown here is, for
example, different scenarios with different bandwidths. For example, the 40 MHz bandwidth will obviously
affect the sensitivity, but at the expense of higher data rates, and also shown is the diversity, 2x2 cases vs.
size source, single input, single output, which ideally, in perfect conditions improve the sensitivity by as much
as 3dB.

Now the blockers are not much of a concern in wireless LAN applications, mainly because it’s a short range
system and we have limited number of users, mostly time division base communicating to one or two access
points, although in recent applications when we integrate the wireless LAN chips along with cellular in a
given cell phone, the cell phone transmitter could impose a fairly challenging docking requirements. But like
I said, mostly what is important in wireless LAN is the noise figure and the signal quality which is EVM and
SNR in the receiver.

18. WLAN Design Challenges

Now, some of the challenges, in the case of wireless LAN like I mentioned in the receiver, the IQ imbalance
and phase noise are important to establish a certain EVM, the same thing with transmitter, usually good
handstand, good systems target for -28dB EVM, in the case of 64QAM, so this is fairly challenging numbers
for as the phase noise or IQ imbalance is concerned, which will affect the constellation as shown here.

The in-band linearity to some degree is important, because it’s also going to affect the SNR. But like I said,
the outer-band linearity is not as much of a concern, unlike the cellular, because of the nature of the wireless
LAN systems.

19. GSM/EDGE

So now we’re going to move the GSM/EDGE system. GSM supports 270 kbps using a fairly simple
modulation scheme of GMSK, this is constant envelope modulation, as opposed to EDGE, which is the GSM
evolution and it supports 3 times higher data rate through establishing a more complex modulation of 8PSK
which is now amplitude modulated as well, and it’s going to impose some linearity requirements in the
receiver.

The bandwidth in both cases is 200kHz and GSM and EDGE are worldwide are defined in four different
bands, the uplink, the receiver vs. transmitter, the downlink vs. uplink has different frequencies so it’s a FDD
system, but also is TDMA meaning that the receiver and transmitter are not operating concurrently, they take
turns and there are slots and frames associated with each, and it’s kind of summarized in this table for your
reference.

Most of the advanced handsets are capable of supporting all four bands of operation, typically GSM 850 and
PCS is used in the US, as opposed to E-GSM and DCS which are more common in Asia and Europe.

20. GSM Noise Figure

Now, in the case of GSM, the sensitivity is defined to -102dBm, that’s again a number coming from the
standard, it’s out of our control, as far as RF designers are concerned. And it’s mainly to guarantee a certain
coverage, given a certain cell or a certain range. Now we can use our equation we had early on. The
bandwidth in the case of GSM, like I said is 200 kHz roughly, the SNR in this example I’m assuming is 6dB
although it can be better, so that combined, gives us about 59dB. This is mostly out of our control. Again, a
good modem can establish somewhat a better SNR, but more or less, it’s a given number.

But it’s noteworthy, most of advanced handsets target for -109 and -110dB, about 7, 8dB better than what
standard requirements requires, mostly to establish a better phone call quality or wider range, or better
reception.

So assuming we have 3dB loss at front-end which is the filters and switch and everything associated there, it
translates to a noise figure of between 2-3dB in the receiver, which is a fairly challenging number. But what
makes it more challenging in the case of GSM

21. GSM In-Band Blocking Requirements

is the fairly stringent blocking requirements. So like we talked about, we have the reciprocal mixing scenarios,
and there’s usually a blocker profile associated for any given standard, in this case, the GSM. The closer the
blocker is in terms of frequency offset, with respect to the desirable signal, usually the less or the weaker it is
as we move further out it gets stronger, it mainly has to do with the way the frequencies are assigned to
adjacent bay stations.

Now the desirable signal is defined to be 3dB above sensitivity, that’s -99, and this is a blocker profile that
starts from 600kHz all the way to 3MHz and beyond which flattens out here.

Now as far as the reciprocal mixing is concerned, we can simply work out the numbers, the desirable signal is
-99, the blocker is also given here, this number is fixed, so that simply translates to a given number. Now the
most stringent cases are 600kHz and 3 MHz blockers which are given here, and again we can work this out
and come up with phase noise numbers, but note that these phase noise numbers assume everything else in
the receiver is ideal. In practice we have others sources of nonideality so typically we tend to design for better
phase noise than the numbers given here.

Now, in addition to the reciprocal mixing, we also have to be worried about the compression, mostly for the
3MHz blocker which could be as strong as -23dBm. Remember that the receiver front-end could have as
much as 30-20dB of gain for obvious noise concern and a 3MHz away blocker is subject to little or no
filtering at the front-end so this establishes fairly stringent compression requirement which will add further to
our budget.

22. GSM Blocker NF

Now to give you an example here, as to how these numbers work out, so in the previous slide,

23. GSM In-Band Blocking Requirements

Remember I said if everything is ideal for in the case of 3 MHz blocker to pass the standard with 0dB of
margin, we need -136dBc/Hz phase noise, but in reality,

24. GSM Blocker NF

Because of the other sources of imperfection, we should target for something better. So let’s say we start
with -140dBc/Hz, 40dB better than the number I gave you in the previous slide, at 3 MHz away. Remember
we had this equation which gives the receiver a blocker noise figure, that is a noise figure when we have a
blocker present. And this number is given that’s -26dBm, the phase noise we assume to be -140. So the
combined effective noise figure is 8dB.
Now a receiver has its own thermal noise floor, so in this example, we assume its 3dB and an additional 3dB
front-end loss, which makes it 6dB. So the combined noise figure will be 10dB. Now let’s see what is
required as far as the standard is concerned. So again we have our noise figure equation which was given
earlier, this 59 is the bandwidth plus SNR, it’s a fixed number, 174 is also given. We also said that the signal
is at -99 dBm, that’s 3dB above the reference sensitivity, assuming 3dB of loss at the front-end, the noise
figure at the LNA was calculated to be 13dB. So this is the requirement which is merely coming from the
standard

25. GSM In-Band Blocking Requirements

which is mainly defined by this blocker profile and the signal vs. blocker relative strength or weakness,

26. GSM Blocker NF

while we can actually come up with phase noise and noise figure numbers. Now with phase noise, and with
this kind of noise figure, we have about 3dB of margin between what is the absolute needed as far as the
standard is concerned and what we can achieve.

Now what makes things further and hasn’t been calculated here, is the kind of compression that the blocker
can impose to the receiver. So the 3dB noise figure in the receiver we started with, is in the case that we have
a weak desirable signal, and we enjoy from the high fun and game, as it is intended for, but in reality the
blocker can actually compress the front-end to some degree and elevate this noise figure, meaning that this
number will now get closer to this 13dB. So that gives you an idea of how you would work out and do all
these tradeoffs.

Again, the 3MHz blocker is subject to little RF filtering, so the filtering is not going to help us here much, it’s
mainly the linearity and phase noise of the receiver which is driven here.

27. GSM Out-of-Band Blocking Issue

Now, in addition to the in-band blockers which I showed you and like I said, the signal is -99 and at 3MHz
away and beyond, it goes up to -23, we have an outer-band blocking profile. Shown here is an example for
the PCS band, all the other three bands are similar and what it says is that from 80MHz away from the edge
of the band, which is 1990 MHz, the outer-band blocker can be as large as 0dBm.

Now this number is coming from the other random users, not necessarily the GSM systems which are present
out there, and this sets a fairly challenging compression requirement for the receiver, it’s essentially 0dBm at
the input, corresponds to about 1V swing peak to peak, which will heavily compress the receiver and because
of that, we usually use external filters and the 3dB front-end loss that I mentioned earlier, a big portion of it is
coming because of the filter, which is inevitably needed because of this outer-band blocking profile.

28. GSM IIP3 & IIP3 Requirements

So, we did talk about the compression and reciprocal mixing and now we can also have a look at the third
mechanism, the 3rd order or 2nd order intermodulation, in the case of GSM, the two blockers are defined to be
800kHz and 1.6MHz away. Remember that the channel spacing in GSM is 200kHz, so we have potentially a
blocker at every 200kHz away from the desirable signal. Now the linearity, the 3rd order linearity is defined,
based on this scenario. The blockers are assumed to be -49dBm at the antenna, again the signal is 3dB above
the reference sensitivity -102dBm, as I mentioned. We had this equation so we can simply plug the numbers
back in and it gives us an IIP3 of -21dBm.
Again this doesn’t mean that a receiver with -21dBm IIP3 is acceptable, because we have other contributors,
so is the very minimum, assuming the receival is perfect, otherwise, and it’s only suffering from 3rd order
nonlinearity. Yet in reality, we have reciprocal mixing, still, because of these blockers will be down
converted noise LOs because we have the receiver on thermal, noise floor and so on. So we usually target for
much better than this.

An in the case of 2nd order nonlinearity, the standard assumes a blocker at 6MHz away, -31dBm, and again
this equation can be easily calculated. We have 6dB SNR, the same number we’ve been using throughout the
presentation. Signal is at -99, the blocker is at -31, so this is the blocker minus signal plus the SNR minus the
blocker, that translates to + 43dBm of IIP2.

Now this assumes that the desirable signal is exactly at DC. So if we have a receiver which enjoys a higher
IF, or intermediate frequency, this will be less of a concern because this will be subject to some filtering.

29. Receiver SNR

Now, besides the linearity and noise figure which are mainly parameters which establish the receiver quality
at the extreme cases, meaning that the signal is subject to weak desirable signal in conjunction with large
blockers. We also have to worry about cases where we have the strong receive signal and thus not much
worry about thermal noise figure or blockers, but still we want to establish a good reception quality.

So shown here is a realistic example of the SNR, the signal-to-noise-ratio vs. the input. What is needed is, a
wrong sensitivity at this point, we have about 6dB SNR, so -112, that’s kind of the sensitivity we’re getting in
this example, the 3dB front-end loss will make it -109, that’s kind of a typical number. And as the input
grows, the SNR is expected to improve, that is mainly because of the fact that when we have high quality or
strong desirable signals, we would like to enjoy from the maximum throughput. And the SNR is expected to
raise and eventually it will flatten out because of 2nd order parameters which we hadn’t considered so far.
Namely the in-band phase noise or the IQ imbalance or for that matter, in fact the in-band of the linearity of
the receiver will basically give us a flat region in this example about 35dB of SNR which is efficient to enjoy
good throughput when the desirable signal is strong enough or basically residing in this region of operation.

30. GSM/EDGE TX Requirements

Now, back to the transmitter, like I said, we have the mask requirements, in the case of GSM and EDGE the
most stringent requirement is at 400 kHz, the mask starts at 100kHz and goes beyond tens of MHz. But given
that typically, like I mentioned, the mask is set by the phase noise and phase noise typically enjoys a
20dB/decade roll off to a good extent at least. Usually we have one or two key points which set the most
stringent mask requirements and everything else beyond that will be automatically satisfied as we have the
20dB/decade roll off in the phase noise. In case of GMS/EDGE it happens to be 400kHz, the requirements is
-60 or 54dBc respectively.

The transmitter quality is defined in terms of phase error and EVM which is shown here, and also one
stringent aspect of GSM is the noise of transmitter and receive band which starts at 20MHz away. We’ll talk
about that in more details shortly which is given here.

31. Mask & Phase Error Calculations

Now, here is shown a more practical calculation as far as the GSM 400 kHz mask is concerned. If you recall,
I mentioned its 400kHz, that’s the requirements, 60dBc. So let’s say we have 3dB of margin, makes it 63, we
allocate about 2dB for process variation, makes it 65. We can simply translate this number to a phase noise
requirement for a transmitter LO which is shown here. Now, what this equation says, is that relatively the
noise mask is 65dBc, the measurement is done at 30kHz resolution bandwidth, so this number will be then
dBc/Hz by taking the bandwidth portion out. There’s a 9dB adjustment factor because the GSM signal is 200
kHz wide, so at 30kHz resolution bandwidth, the power, the peak power is about 9dB lower than what it
would be in an infinite bandwidth system. So we have this adjustment factor and this translates to a fairly
stringent phase noise requirement of close to -119dB/Hz at 400kHz. So this is an example of how the simple
mask number is calculated back to a more tangible or friendly number as being the phase noise in this case at
this 400kHz offset.

Now, like I said, the in-band phase noise is also important for establishing a good transmitter quality. Now I
have made a fairly simplistic scenario that we have a given in-band phase noise, in this case -89dBc/Hz which
is flat up to 100kHz, that’s the GSM bandwidth. Now, this number integrated over 200kHz will give us
-36dBc. Now this corresponds to probably .02 radiant which is 1°.

Now recall that the phase error needed was 5°, so let’s say we have 2° of margin, and another degree assigned
to process variation, so that makes it 2, so simply the in-band phase noise of -89dBc/Hz or -36dBc integrated,
will establish half of that budget of 2° which we have. So 1° is simply coming from this, and then there are
other contributors such as the modulator itself, the filtering, the group delay and so on which will add to this
and make it closer to the 2° budget.

So with these two pieces of information, we can simply define a phase noise profile for the transmitter, this
being the in-band and this being the 400kHz and beyond.

32. GSM RX-Band Noise Requirements

Now, like I said, for the case of GSM, we go all the way to 20MHz offset, that’s driven from the fairly
stringent requirement of having a transmitter noise of better than -79dBm in the receive band. The idea here
is that if you’re transmitting at full power, which in the case of GSM is +33dBm at the antenna. If you
happen to be at the edge of the transmit band, in this case 915MHz, then the tail or the phase noise
transmitter will establish a noise floor to a nearby handset which happens to be tuned at the edge of the
receive band, in this case being 935MHz or 20MHz apart. Now this is simply again, a very simple phase
noise calculation, we start with -79dBm, the measurement is done at 100kHz bandwidth, so we make it
dBm/Hz, and then relative to the power, makes it dBc/Hz and that’s the number here. Now this is for the
entire system at the antenna, the PA has its own noise floor, so our target was -79dBm at the antenna, let’s
say the PA is -84, 83, these are typical numbers, so that leaves a fairly stringent noise requirement of -166 for
the RF IC. So this way, we have the in-band phase noise of the transmitter, -89dBc that I showed you, the
400 kHz -19, and it continues all the way to -166.

The problem here is that as you all know, the phase noise flattens, it doesn’t enjoy from the 20dB/decade roll
off because of the buffers and the power amplifier, power amplifier driver, so this puts fairly stringent
requirements on the entire transmitter and not just the VCO which is mainly the case for the 400kHz.

33. Introduction to W-CDMA

Now, like I said, to establish higher data rate, at some point the wide band CDMA was introduced. This is
also an FTD system, but is based on this spread spectrum. It’s a code multiple access, meaning that the bit
rate is set but also there’s a chip rate, meaning that the desirable signal is affectively multiplied by high
frequency code and it’s spread and the opposite thing happened in the receive end, the signal is despread. So
ideally, it gives us some advantage in terms of more resilience to outer-band blockers, because the blocker is
subject to spreading at the receiver end, as opposed to the desirable signal which is subject to despreading, so
the energy kind of changes once we filter this portion.
Now currently wideband CDMA supports up to eleven bands, a lot of them overlap with the four GSMA
bands that I showed you earlier,

34. WCDMA Full-Duplex Problem

Although it has it has some of is own new dedicated bandwidths, and handsets selectively support some
bandwidths depending on what kind of region they are targeting.

Now in the case of the wideband CDMA, it’s a full-duplex system, meaning that the receiving transmitters are
operating concurrently. So the blocker is mainly associated with the own transmitter as long as the receiver is
concerned. So shown here, while the receiving transmitter are concurrently operating, ideally the duplexer,
which is a dual-band filter will separate them at the antenna, but in practice, there’s a final leakage from this
band to that to the receiver band, so part of this shown TX signal, leaks back to the receiver, which is also
receiving at the same time. Now this sets two problems.

One is the phase noise, the far out phase noise establishing the noise floor, which adds to that receiver own
noise, so that’s one problem. And that’s mainly defines the phase noise of the transmitter and the amount of
filtering which can be established at the transmitter output and that’s mainly a cost driven factor. As well as
the receiver linearity, or 3rd order linearity, which for example, it could be that we have a large out of band
blocker at half the space between RX and TX which are at different separate bands like I said earlier. And
because of the 3rd order, we get some kind of inter-moding here and that establishes the IIP3 along with what
kind of filtering can be achieved, prior to getting to the LNA.

35. RX NF Requirement

So, we’re going to talk about those numbers in more details. But before that, we start with the noise figure for
the receiver. Now, in the case of wideband CDMA, that dedicated physical channel energy is defined to be
-117dBm. Note that the bandwidth is 3.84MHz as opposed to the 200kHz for the GSM/EDGE. This
particular measurement is done at 12.2kbps. The chip rate or the spreading factor, the chip rate was 3.84MHz
if you recall, so the spreading factor is 128 or 21dB. We have about 4 dB of coding gain, so this puts this
-117 to here. We need 7dB of SNR, wideband CDMA is QPSK modulation, it requires 7dB. So that’s
combined 18dB, so that brings -117 to this -99, and the noise floor is -174 plus 10log of the bandwidth which
is 3.84, that happens to be -108, so essentially the delta between this noise floor and this is the noise figure, in
this case being 9dB. Now the 9dB is the absolute very worse noise figure we can have at the antenna, so to
that, we have to add the front-end loss, typically 3-4dB. And in the case of wideband CDMA receivers, the
other factor, unlikely GSM, which are contributing to this. That’s mainly because of the fact that transmitters
are on at the same time. So the transmitter noise floor, like I mentioned, will add to that. As well as the 2nd
order nonlinearity of the receiver, what amplitude demodulate the transmitter leakage and further add to this
budget.

So with these reasons, most of receivers have a thermal noise figure of 2-3dB like GSM. So give enough
margin for everything else added to that.

36. IIP2 Requirement

Now, like I mentioned, one stringent problem is the IIP2, not because of an outer-band blocker which was the
case in the GSM, but was because of our own transmitter. The transmitter leakage will be demodulated
amplitude, demodulated and causes spectrum around zero IF which is shown here.

Now, it’s a fairly complex calculation, it’s hard to put it in terms of equation, but system simulation shows
that usually you need better than +45dBm of input IIP2, to make sure that this demodulated spectrum
originated from transmitter is not substantially effecting the receiver noise figure. Now coincidentally, this
number turns out to be fairly close to what we calculated for the GSM IIP2, but because of completely
different mechanisms.

37. Out-of-Band IIP3 Requirements

Now, the IIP3, like I mentioned, has to do with the fact that the transmitter signal leaks back to the receiver to
a finite duplexer isolation, so transmitter at the antenna is +24dBm, that’s the requirement. Let’s assume we
have 3dB of loss in the duplexer and front-end, so that puts the transmitter to a +27dBm. Now, let’s assume
that conservatively we have 45dB of isolation from here to here, so this makes the TX signal to be this large,
then the blocker, according to standard, is defined to be -15dBm at half the duplex frequency, sitting here.

Now this blocker is also subject to some filtering because of the fact that the duplexer at the end of the day, is
a simple dual band filter. So in this case, I’m assuming 30dB of filtering, that’s again a typical number. The
SNR is given here, remember we had 7dB SNR, 21dB spreading factor on 40 decoding gain, so it’s -18, that’s
because of the spread spectrum nature, the SNR is negative in the case of wideband CDMA. Now the
sensitivity is defined to be 3dB above the bare minimum -117. That gives you the IIP3 to be -6dBm.

This assumes, no extra filtering besides what you achieve in the duplexer, so obviously inserting a filter here,
at the expense of cost will proportionally relax this number. Now this number is fairly challenging, and has to
be met at the same time as meeting a good noise figure of 3dB and less.

38. WCDMA Transmitter Requirements

Now, for the transmitter, we have a similar requirement as we talked about in the GSM. One is the far out
noise, like I said in this case, it’s not affecting a nearby user, but it’s affecting the very own receiver where
the transmitter is associated with.

So let’s assume we have 160dBc/Hz noise at the transmit receive offset frequency. In the case of low-band,
that’s 45 MHz, so we are at 27dBm. So remember that that was the transmitter output here. We assume
50dB of duplexer isolation in this example, with 160dBc/Hz, we arrive at a noise floor of -183dBm/Hz, that’s
9dB below -174 noise floor, so this means that it will affect the receiver noise figure some, in this case by less
than a dB. So this will add further to our budget.

Then we have adjacent channel leakage ratios, the most stringent one is at 5MHz set to be 33dBc and of
course the EVM, or the transmitter quality.

39. WCDMA TX EVM

Now, to give you some ideas, the EVM, there are several contributors depending on the transmitter
architecture, in the case of linear transmitters, IQ imbalance, the LO feedthrough, as well as the phase noise,
are typical contributors. So we can have a simple 1st order equation, assuming we have 40dBc IQ imbalance,
40dBc LOFT, and 40dBc integrated phase noise, that corresponds to 1.7% EVM which is simply given here.

And the filter response will add further to that. Most of the handsets are RF ICs target for 3% and better
which is mainly set by how we budget these numbers.

40. WCDMA TX ACLR

Now, the ACLR is shown here. It can be easily calculated backward to an IM3. The wideband CDMA signal
is amplitude modulated so it can be thought of as several signals present at the same time. Each of them, due
to the 3rd order nonlinearity, the receiver will cause IM3 tones which are shown here. Now the ACLR is
measured at 5MHz at ±1.9MHz filter bandwidth, that’s 3.84MHz number, so we can do system simulations
which correlate the IM3 to ACLR. So for example, if we need 40dB of ACLR, we need about 31dB of IM3
so it’s 40 minus this 9dB here.

41. WCDMA ACLR1 Calculations

So an example is shown here, like I said, the standard requires you to have 33dBc. The PA is contributing,
the PA is not perfectly linear and to enhance the efficiency, it’s usually operating close to its 1dB
compression, so let’s say the PA is 37dBc. Now with 2dB margin, we arrive at 40dBc requirement for RF IC.

42. WCDMA TX ACLR

Now, if you look at the chart here, the 40dB ACLR, means that the IM3 is about 31dBc which is -40 plus 9.

43. WCDMA ACLR1 Calculations

So, that means, what kind of linearity we need, so let’s say we are targeting 0dBm output power, there are
two tones, the IM3 is 31dBc, so the output IP3 is simply calculated to be +15.5dBm. That’s about 5.5dBm on
1dB compression. So we are about 3dB below 1dB compression to achieve this kind of ACLR.

44. Receiver Architectures

So, now we’re going to talk about the transceiver architectures based on what is required as far as the
standard is concerned.

45. Ideal Receiver

So ideally, this is kind of the dream receiver which was at some point, proposed by Motorola, back in twenty
some years ago. It’s simply a data converter sitting at a front-end and everything will be directly digitized and
processed in the back-end comfortably through some strong DSP unit.

The problem with this simplistic approach is the blocker issue, which we established earlier. We have the
desirable signal, say in the case of GSM at -99dBm, and several in-band and out of band blockers, which
could be as much as 100dB larger. So this sets fairly unrealistic dynamic bench for the ADC, at least as far
this particular type of application is concerned. So for these reasons, most of the receivers today, depend on
this kind of architecture

46. Super-Heterodyne Receiver

And we use means of down conversion, meaning that large portion of the signal processing is deferred to at
significantly lower frequency, in this case IF or intermediate frequency, so this is a super-heterodyne which
was defined many years ago, probably in forties, and what it proposed is down converting the signal as soon
as we can to some comfortably low enough intermediate frequency where we can apply the amplification, the
filtering and further the digitization more comfortably.

47. Image & Half-IF Blockers

Now as you all know, this suffers from image and several other blocker issues. I’m not going to go through
that, it’s a fairly well-known fact, so because of this stringent image and half-IF and other kind of blockers
requirements, it’s not been used because of the fact that we need expensive external filters.

48. Zero-IF Receiver

So instead, typically our receivers today are either 0 IF which is shown here, and this immediately down
converts the signal at the DC. So, it doesn’t suffer from the extensive external filtering which is needed in the
case of super-heterodyne, but it puts the signal closer to DC and thus, more sensitive to 1/f noise DC offset,
and a 2nd order linearity which I showed you earlier, and remember I said the 43dBm number is if the signal is
at 0 IF, if the signal happens to be at a higher IF, this number is proportionally relaxed. So this is a tradeoff
you’re facing here.

49. Image Rejection in Zero-IF Receivers

Now, the image issue goes away in the direct conversion receiver, but it’s not quite true, it’s the image is not
because of an outer-band blocker, but it’s because of the signal own image after the complex down
conversion. So the signal folds and establishes a noise floor which is set by the IQ imbalance in the receiver
and this could be a big concern in the applications that we need high SNR because of establishing a good
throughput.

A good example is, for example .11n or recently .11a/c which is targeting fairly large throughputs of hundreds
of Mbps at the expense of very large SNRs of 30dB and beyond. This will be fairly established, challenging
requirements in the IQ, not because of outer-band image blockers but because of the signal folding itself.

But more or less, people have figured out many ways of dealing with these problems I mentioned, some of it
will be covered later on throughout this short course, so I’m not going to go through the details.

50. Low-IF Receivers

And that’s kind of the architecture of the choice, direct conversion in most advanced handsets or the low IF
which is a compromise between super-heterodyne and 0 IF. It sets its signal at reasonably low IF where the
blockers are in-band and therefore even though the image is a concern, the amount of image rejection needed,
is not as stringent as what it would be in the case of super-heterodyne receiver. So these are mostly common
in pretty much every wireless LAN or cellular handset, either low-IF or 0-IF.

Typically low-IF is more suitable for narrow band applications such as GSM and EDGE, or we can afford
having a sort of bandpass type filter wherefore the wider band application such as 3G or wireless LAN
OFDM, simply use OIF.

51. Choice of IF

Now, in the case of low-IF and particularly the GSM, the choice of IF is of interest. There is a tradeoff, the
higher the IF, the less sensitive we are to known problems of 1/f noise or DC offset, but the image rejection is
more challenging.

So shown here is the adjacent blocker profile which is similar to what I showed you earlier for the case of
in-band blockers, and as you can see and as expected, the blocker dramatically increases in terms of the strain
as we move further along. So the higher the IF, means that the stronger the blocker will be, and
proportionally the image rejection dramatically increases. So for the low-IF of say 100kHz, we have a fairly
manageable less than 30dB image rejection requirement. Whereas if we moved the IF, for example to
200kHz, to put the signal further away from DC offset to other known issues, the image rejection goes up to
something which typically cannot be achieved unless some complex calibration is used in the receiver.

52. Dual-Conversion Receiver

Now, another compromise is dual-conversion receiver which we do two steps of down conversion. The
advantage here is that the second down conversion is at a lower frequency, and thus, the typical problems are
0-IF which are mostly related to the LO quality or the LO frequency are less stringent. But again, it does
have the image issue, and in general, the blockers are more problematic here, so most of the handsets are
receivers today and again, don’t use this architecture, although it’s common in certain applications.

53. ADC Requirements

Now, before we finish this receiver section, we’re going to briefly talk about the ADCs, so for a given
receiver, 0-IF or low-IF, we already established the entire chain requirements, as far as phase noise or noise
figure, out of band and in-band linearity is concerned, so add the numbers I gave you is for the entire
receiver, we can easily allocate it for a given blocker using known equations which show the noise and
linearity for a given gain for a given block. What is left here, and we haven’t talked much about, is the
filtering. So, like I said, a lot of the aspects of receivers, particularly those which are related to blockers, are
substantially affected by the kind of filtering that can be achieved. Again, the external filtering is mostly for
out of band blockers, comes at a cost. Integrated filtering usually happening at low frequencies and this case,
0 or low-IF, will target the in-band blockers.

Now at the end of the day, there is a tight compromise between the filtering we achieve at IF and the kind of
ADC we will need. So the ADC has to have several requirements satisfied, first of all, its quantization noise
floor should be well above that of the receiver noise floor, not to degrade the noise figure sensitivity
significantly. More importantly, it also is a question of what kind of filtering can be achieved in the receiver
and what kind of blockers suppression is achieved, prior to hitting the ADC. And that sets the dynamic
range. So there’s a tradeoff between the cost, which is the size of filters here, as opposed to ADC dynamic
range.

54. ADC and Filtering

And I have shown you an example here to kind of make this problem more clearly. First of all, let’s say we
have, a desirable signal that’s weak at the input, it’s accompanied by some large in-band blocker. So shown
in green, is the signal flow which is going through several stages of amplification and filtering, there is some
art-of-filtering, and some IF amplification and filtering, so the signal proportionally increases. And by the
time it hits the ADC, it has a certain level which is shown here at this point, which is here.

Now shown here, is also the same progression for the blocker. Now, like I said, if the blocker is in-band, it’s
subject to almost no filtering up to this point. So the blocker is similar to the desirable signal amplified, but at
this point, it is subject to some filtering, depending on the order, or the cost, at the end of the day, so it sees
less of a gain or some attenuation and it puts the blocker at this point.

Now in many applications, such as GSM, the desirable signal and blockers can up or down fade. So this
means that, note that the fading is fast and the ADC is not going to respond to that. So we have to allocate
some budget for the potential up or down fading, and this means that the blocker which at this point, is the
strongest signal on entering the ADC subject to some more amplification or some more, being stronger
through fading. So this puts it here. Now this will be the ADC full scale, that’s the biggest signal that it can
tolerate. Now in the other end, the signal is sitting here and we don’t want the ADC quantization noise to be
too close to this signal. So we’ll say we need 10 or 15dB of margin, so this essentially sets the ADC dynamic
range. The more filtering we achieve here means that the weaker it will be and the less dynamic range that
we need for the ADC, at the expense of higher cost of building good filters.

55. Receiver Gain Control

And also there is obviously some gain control associated with the receiver which is shown here. Again, if we
have only desirable RF signal, then we can adjust the receiver gain in stages proportionally to avoid
compression, but then in often cases, we have weak signals, accompanied with large blockers and ADC
usually has to respond to the weak signal and it may not set the receiver gain as it’s needed as far as the
blocker is concerned, so hence the compression, could be an issue, although we can have some provisions
here.

For example, we can have a narrow RSSI which monitors the signal energy after the filtering, so that’s mainly
the desirable signal, as well as some wide-band RSSI at which responds to the signal energy prior to filtering,
and this case blocker, and we can allocate some kind of different gain setting which puts us in a more
favorable spot as far as the noise figure and the blocking requirements are concerned.

56. Transmitter Architectures

So, now let’s move on to the transmitter.

57. Direct-Conversion Transmitters

Like the receiver, the most common transmitter architecture used today is direct-conversion, which is
versatile, it can support any modulation. It’s fairly low powered highly integrated, but suffers from two
distinct problems. One is pulling, meaning that the PA modulated spectrum is exactly at the same frequency
of the VCO single tone and it could affect it adversely.

And the second problem is the fact that this receiver, this transmitter has potentially worse noise response
which is a key concern when we’re dealing with GSM and 3G systems like I showed you earlier. The far out
noise is certainly a big concern here.

58. Dual-Conversion Transmitters

We could also talk about dual-conversion transmitter, very similar to the receiver. Again, it kind of relaxes
some of the requirements because of a lower frequency up conversion here, although this is not common at
all.

59. Linear TX vs. Translational Loop

What I mentioned earlier, the fact that we have a very stringent 20MHz noise issue, remember that
-166dBc/Hz number we calculated because of the receive band noise, and in case of 3G -160. This could rule
out a direct conversion transmitter here in the case of GSM, mainly because in this case, not only the VCO
and LO contribute to this 20MHz far out noise, but the entire transmitter chain, the modulator that DAC in
the filters are also contributors.

So for this reason, majority of GSM transmitter throughout decades ago have been using translational loops
which is essentially a offset PLL, translating the spectrum at IF to appear exactly at the antenna through a
PLL. Now the advantage here is that the noise of the mixer, as well as the loop components are subject to the
loop filtering, so the only contributor here, is the VCO and the LO chain, as opposed to this. So this is more
of an architecture of the choice in case of GSM, because of the 20MHz noise issue.

And other advantages, the pulling is relaxed because the VCO is modulated exactly the same way as the PA
is, unlike it here. The problem though is, it only supports envelope or phase modulated standards which is the
case for GSM.

60. PLL-Based Transmitters

It can be extended or simplified to a simple delta sigma PLL, instead of having a translational loop which
requires a mixer and a second LO, we could just have a PLL, using delta sigma modulator to provide the
modulation. It’s essentially the same type of transmitter benefits from the same low noise features of
translational loop, but it’s more complex because it’s more difficult to design a narrow band PLL at good
phase noise requirements.

61. Basics of Polar Transmitters

Now, like I said in both cases of translational loop and PLLs are only suitable for GSM, in case we’re using
EDGE which is using HPSK and some amplitude modulation. It can be extended to a polar architecture.
Which we have the same PLL, establishing the phase combined with an alias path which is providing our
amplitude, and once they are combined, they give us the composite spectrum which is intended for the case
of EDGE.

This is obviously a lot more complex architecture and suffers from several drawbacks, we’re going to briefly
talk about that.

62. EDGE AM & PM Signals Spectrum

So in order to understand the tradeoffs, what is shown here is the final 8PSK spectrum in the case of EDGE,
that is

63. Basics of Polar Transmitters

what’s going to appear here, which is the combined spectrum vs. the phase and amplitude standalone
spectrums.

64. EDGE AM & PM Signals Spectrum

So as you can see, particularly the phase signal is quite wideband and even though the GSM signal at about
400kHz, the way it drops by more than 7dB ideally, the phase is quite wide, up to over 800kHz. So this
means, that in order to successfully support the EDGE to a PLL, we need a fairly wideband PLL to
successfully pass through this wide spectrum or it requires some other tricks which I’m going to talk about
later, shortly.

65. EDGE AM & PM Delay Mismatch

Also another concern is the alignment between AM and PM. Again this is a well-known figure, and as you
can see, what is shown here, is the spectrum vs. a potential delay which can be introduced anywhere in a
signal path between the AM and PM signals. So far we have assumed they are ideally combined and they
give us the intended spectrum which is in black here. What if AM is delayed with respect to PM because of
the fact that we have filters in the PLL and the AM path? They’re subject to variation and hence, there is a
combined group delay uncertainty that will lead to some spectrum expansion. So for example, in case of
EDGE, remember where we had 54dBc requirement for the entire chain, let’s say, the RF IC needs 60dB or
better, here, so this sets the delay to be 40ns or better, which is fairly challenging.

66. AM & PM Path Non-Idealities

In order to understand the challenges here, let’s take a closer look at our polar transmitter. So what we have
is several contributors inside the PLL, that’s our PM path, as well as the AM path.

Now typically, we use narrow-band PLLs for obvious noise reasons, so this means that we need to have some
kind of predistortion to overcome the bandwidth problem and the mismatch between the digital predistortion
block, as well as the loop filter characteristics or the PLL characteristics which are analog and subject to
variations will give us basically some spectrum mask violation shown here.

Like for example, 4% mismatch, gives you 66dBc. Now, there are other contributors. We have the phase
feed-through, meaning that part of the PLL output can feed through without being subject to combining at the
output and given that the phase signal is pretty wide, it will establish a noise floor, in this case 65dBc. We
have the linearity of the loop.

Again typical numbers can give us something like 66dBc of 400kHz mask, as well as the phase noise.
Remember, we actually calculated this number at the beginning of the presentation. For 65dBc 400kHz
mask, we ended up with minus, close to 19.

And then we have the AM path, non-idealities which is mainly coming from the nonlinearity of the PA driver
or the PA, mainly the AM-AM and AM-PM. So they also contribute to mask so you can see that unlike the
GSM, which is mostly set by the phase noise, there are several factors which combine. In fact, in this case,
we add them up, going to arrive at 58dBc, which doesn’t leave us a lot of margin to that 54dBc requirement.

67. Polar Design Trade-offs

Now shown here, is kind of a more graphical illustration of the tradeoffs that I mentioned to you earlier. So,
the most important one is the PLL bandwidth tradeoff for noise reasons we would like to have a narrow PLL,
say 200kHz in this example. Now remember that the EDGE spectrum is 800kHz wide, so it needs a
predistortion, and for that to be successful, it requires a fairly exact estimation of the PLL frequency
response. So we need some kind of calibration here. Then we have the AM-AM and AM-PM problems as I
mentioned. We can linearize the PA driver, simply by burning more current.

One good feature of polar is the fact that it can drive the blocker closer to saturation, so that the more
linearity you need, it corresponds to less efficiency.

68. Next Generation Cellular Transceivers

Alright, so I’m going to basically finish the presentation by giving you a kind of highlight of what is recently,
basically, coming up.

69. EDGE Evolution

So, we did talk about GSM, EDGE and wideband CDMA in details in the previous section of the talk.

There are also some enhancement again, as I showed you earlier to establish higher throughput. So one is the
EDGE evolution, the EDGE is 8-PSK modulation spectrum, I’m sorry, it supports a data rate at about 800kb/s
and that’s not the throughput, that’s just the bare data, so the throughput is less than that.

EDGE evolution supports a more complex modulation in a scheme of 32 QAM, at the same 200kHz
bandwidth for the downlink receiver and 16 QAM for the uplink, instead of 8-PSK so this essentially means
more SNR, more stringent linearity and noise requirements, but at the expense of say 2-4X higher
throughput.

It also relies on diversity, the concept of MIMO which I mentioned earlier, it started in wireless LAN, but a
lot of cellular handsets today are also targeting diversity for the same reasons as we have in the wireless LAN.

So this puts us in the throughputs of closer to 1Mb/s as opposed to what we used to have in the case of
EDGE.

70. LTE Features

Now, we have also LTE which is a long term evolution, and in the case of 3G, we had about 5MHz,
bandwidth 3.84 to be exact. The LTE support up to 20MHz, so that’s about 4-5X wider band, that essentially
means proportionally higher throughput at expense of obviously more complexity. So at the end of the day, it
can support up to about 100Mb/s data rate for the case of wideband CDMA, we were about 1 or 2Mb/s so
this is dramatic enhancement by using much more complex modulation scheme, say 64 QAM as opposed to
QVSK, as well as wider bandwidth and also introducing diversity.

The LTE could support full duplex, similar to 3G, so the TX leakage and all that is obviously a concern, or it
could be TDD system, or TDMA system, like GSM, which relaxes some of the problems we had.

One advantage, important feature of LTE, if you want to compare it to WiMAX, which also has similar type
of modulation scheme and speed, is the mobility, which is up to the speed of 120km/h can be supportive with
high performance. It`s obviously backward compatible with GSMA.

71. Peak to Average Ration for 20M LTE TX

As an example, because of the lack of time, I can`t go through the very details, but I`ll just show you a few
features. So, one is for example, the peak to average ratio or linearity in the transmitter, how does that
compare to 3G which we calculated in details?

For the case in 20MHz LTE, that`s the wider bandwidth, shown here is the peak to average ratios. Like
wireless LAN, it`s adaptive, meaning that the better reception we have, the more complex modulation we will
use and hence, more data rate. So it can be anything from QPSK to 64 QAM, and we`re looking at about
6dB of peak to average ratios. It means about a few dBs, a little bit more stringent linearity, compared to
wideband CDMA, but the ACLR requirements is 3dB relaxed.

The 3G was 33dBc if you recall, we went through that in details. LTE is 30 and the output power is about
1dB lower so at the end of the day, you arrive, surprisingly at similar linearity requirements, compared to
wideband CDMA, as far as a direct conversion transmitter is concerned.

72. Impact of TX Phase Imbalance

Now similar to the 3G case which I showed you the simple equation, the IQ gain and phase imbalance, as well
as phase noise effects the modulation quality, so again it`s hard to put these in simple equations and it usually
requires a complex system simulation so an example is shown here, so this an ideal, 64 QAM output, in the
case of LTE, and the EVM is less than 1%, it`s mainly set because of the filtering and modulator quality and
nothing else. Once we start introducing more and more IQ imbalance, the consolation gets corrupted and
then we arrive at let`s say, 3.6% for 4°, which is not a dramatic degradation, but again, it sets some
requirements as far as IQ gain and phasing balance is concerned,

73. Impact of TX Gain

again, imbalance is actually shown here. And again, similarly, up to 1dB can give us 4 1/2 % of dBm as
opposed to an ideal case.

74. Impact of TX Phase Noise

Now in addition to that, we have the in-band phase noise, the same as 3G, this is the ideal case, and these are
the simulations with two phase noise profiles which are shown here. I think these are actually swapped, so
this is profile 2, which is this, and this is profile 1, so this is the worst phase, so please correct that.

And you can see that this kind of phase noise, which is not necessarily too easy to achieve, could contribute
to as much as 8% of dBm, or so.

So, basically, what we have is the standard which is a high level document and what we do is based on with
simple hand calculations I showed you, and also more complex system simulations like the one I`m showing
you. We can translate those high level numbers to more friendly numbers as far as RF design is concerned
that is essentially in-band phase noise, IQ imbalance, noise figure linearity and so on, and then we can budget
the entire receiver and transmitter, based on the tradeoffs we have.

So I hope that throughout this presentation, I have been able to give you an idea of how this whole process is
being done, and it`s a fairly cooperated process between the system designers who start with the high level
standards and translate it to these numbers. And the circuit designer, as well as what is achievable at a
reasonable cost, for a given receiver.

And a lot of these, like I said and showed you, are a strong function, of what kind of receiver or transmitter
architecture we`re using. A given standard will certainly require different architecture to relax some of these
requirements.

75. Handset Calibrations

So the last thing I`m going to talk very briefly a few more minutes,

76. TX Automatic Calibration

Is the handset calibration. So, we have the requirements as far as, for example, when we`re talking about
transmitter, I showed you how important IQ or LO feed-thru is for example, to give us a good dBm, or the
linearity and so on. So, a lot of them cannot be guaranteed by design, or if so, it will come at a fairly high cost
of say, much bigger area, much higher power consumption. So often times we let our transmitter receiver to
be not as great as we`d like it to be, and enhance it by using some automatic calibration schemes.

So an example is shown here, for example, for the case of IQ and LO feed-thru which as we established
before, is directly impacting our dBm, if you have a perfect transmitter with no imbalance, and let`s say we
transmit a single tone at baseband, which will be up converted to a tone at RF here, the consolation, the I vs.
Q will be a simple circle, now because of the IQ imbalance, this is not going to be a circle anymore, or it will
kind of look like an oval, and because of the LO feed-thru, its sender will shift.
So one fairly simple trick to overcome this IQ or LO feed-thru problem is to have a peak detector at this point
and essentially envelope detect what we have here, which will correspond to this red signal shown here. In
the ideal case, we have a constant flat signal, is just a tone, which is being envelope detected, so it`s a
constant envelope, but because of having LO feed-thru and image, we have a break frequency which is set by
the delta between the image and the desirable signal. And we can monitor this, this is a very low frequency
signal to achieve ADC and just for example, the DAC, IQ imbalance to essentially take this away and make it
more like a perfect line.

So this allows us to design, say a lousy transmitter, which means at a good cost, or low power consumption,
but enhance it through this.

Then we had the filter, for example, the loop filter calibration for the case of PLO base transmitter, remember
in polar, I showed you, it`s important, so we could use other calibrations like that and also some power
control, which means we could compensate for the gain variation of the transmitter, which is typically hard to
stabilize perfectly over process, by again using peak detectors.

77. RX Automatic Calibration

We could use similar auto calibrations in the receiver. In the case of low IF receivers, remember I showed
you because of the in-band blockers in the image, depending on the IF frequency, we need fairly stringent
image rejection requirements, so we can do digital enhancements for the IQ and in the case of 0 IF, we need a
good IQ to achieve a good SNR. So again, that can be calibrated, same thing with IIP2, there are many
examples of designing a receiver with lousy 2nd order linearity and enhanced through calibration.

One example was shown here, which you apply test tone through the transmitter, your own transmitter, which
is AM, amplitude demodulated here, and try to minimize that by enhancing the 2nd order nonlinearity.

78. Factory Calibration

And finally, for most of the handsets, we have inevitably factory calibration, there’re usually three categories,
one is the crystal oscillator, this varies quite a bit over process, over 10ppm or so.

In the case of GSM, we need better than 0.1ppm, very stringent accuracy in the final frequency, so inevitably,
we perform some crystal calibration, meaning that we tune the crystal oscillator. We also need an exact
receiver gain. If you look at your cellphone, the bars you get, essentially which shows what reception you
have, is through an RSSI. And this needs to be accurate within ±2dB. So because of the gain variation in a
receiver which is inevitable, we may perform some factory calibration, and the same thing with a TX,
although we could use some envelope detector and some power control like I showed you earlier.

So this, more or less, concludes my presentation. Basically, what I tried to do, as I said was, to start with a
high level standard, and see how we can translate it to more friendly numbers for the RF design. Through
some hand calculations, some intuition and some fairly complicated system simulations.

So this is it, thank you.