CX74017

On the Direct Conversion Receiver

6

Skyworks Solutions, Inc., Proprietary and Confidential

101735A

Preliminary Data Subject to Change

July 20, 2001

In TDMA receivers that cannot be AC coupled, the idle time slot,

that is, just before reception, can still be put to good use by

storing the value of the offset on a capacitor and then

subtracting it from the signal path during the burst. This is

exactly the same method which is normally used to correct DC

offsets occurring at the second mix of superheterodyne TDMA

receivers, where this mix goes to baseband. In that case the

only problem causing DC is LO self-mixing. In this method, the

DC value produced by the receiver is obtained in a pre-

measurement prior to the receive burst.

It is important when using this method, that the signal path prior

to the mixer be opened during the DC pre-measurement, to

prevent any large blocking signals from affecting the result.

Blocking signals, which can appear at any time, most often

induce variable or wandering offsets. The measurement-and-

subtraction process cannot correct these offsets, because the

blocking signals may appear during the measurement and not

during the burst, or vice-versa. For blocking-induced DC, the

most effective measures are the elimination of self-mixing paths

and the maximizing of linearity to prevent the DC to start. Failing

these, there is still the possibility of DC-correction after-the-fact

in the digital signal processing (DSP) occurring at baseband.

DSP techniques can be used to remove the DC offset in TDMA

systems, in a way that cannot be duplicated in the analog

domain: a full timeslot of the received signal can be buffered,

the mean of which is determined and then removed from each

data point of the signal. The resulting signal has zero mean. For

the signal is lost, but the typical effect of this is minimal.

Figure 12 illustrates the use of such a method for a typical GSM

receiver. This technique can be further refined by tracking the

mean over portions of the burst, allowing the detection of

sudden interferers or blockers and canceling their DC product

only where it occurs.

Careful layout can also improve isolation.

101735A 12_071801

Figure 12. BER Improvement with DSP-Based DC Offset

Cancellation

Non-Linearities

As mentioned previously, another DCR problem is non-linearity.

Just as with the superheterodyne receiver, the DCR exhibits

spurious responses. For the superheterodyne, these occur at

RF input frequencies where

()

LO

M

RF

N

=

±

,

while for the DCR they occur where

()

LO

M

RF

N

=

−

When a blocking signal’s carrier falls on one of these spurious

frequencies, the signal is translated to baseband with an

attendant shift in its bandwidth, dependent on the spurious

order.

However, more importantly, large blocking signals also cause

DC in the DCR, whether on a spurious frequency or not. The DC

is produced at the mixer output and amplified by the baseband

stages. It is due primarily to second order mixer non-linearity,

characterized by IP2 (second order intercept point), IM2 (second

order intermodulation.) It can be alleviated by extremely

well-balanced circuit design. However, only a short time ago, the

mixer and LNA used to require a single-ended design because

the antenna and a hypothetical preselect filter were usually

single-ended.

In most systems, third order intermodulation is important, as it

usually falls in-band, in the vicinity of the signals of interest, and

is characterized by IP3 (third order intercept point). In direct

conversion, the second order intermodulation becomes critical,

as it produces baseband signals, which now appear as

interfering signals in the down-converted desired signal. The

second order non-linearity is measured by the IP2. IP2 is

defined in the same manner as IP3 Figure 13.

Either a 2-tone, or 1-tone test can be performed, and the IP2 is

defined by extrapolating the low-frequency beat tone in the

former or the DC component in the latter, until it intercepts the

fundamental curve. To illustrate in the case of a single tone test,

the input signal is:

)

cos(

)

(

t

A

t

x

ω

=

Assuming a non-linearity modeled by a polynomial:

...

)

(

3

3

2

2

1

+

+

+

=

x

a

x

a

x

a

x

y

...

)

2

cos(

2

)

cos(

2

...

2

1

)

2

cos(

)

cos(

)

(

2

1

2

2

2

2

1

2

+

+

+

=

+













+

+

=

t

A

a

t

A

a

A

a

t

A

a

t

A

a

x

y

l

fundamenta

DC

ω

ω

ω

ω

43

42

1

3

2

1