GC1115_06 Datasheet, PDF (7/82 Pages) Texas Instruments – Crest Factor Reduction Processor

GC1115
www.ti.com
SLWS144C – FEBRUARY 2005 – REVISED JUNE 2006
During peak detection, an effective 256x oversampling process identifies the exact magnitude and location of
interpolated peaks. In other words, the interpolated waveform approximates the output of the follow-on D/A
converter. The interpolated waveform can therefore have a larger amplitude than any of the D/A’s individual
input samples. The GC1115’s effective 256x oversampling allows both the magnitude and the location of peaks
(after D/A conversion) to be predicted prior to the actual D/A conversion process. If the magnitude of the
interpolated peak is above the detection threshold of the PDC stage, the peak is a candidate for peak
cancellation. Detected peaks are always cancelled, unless all of the cancellation resources assigned to a given
stage are already busy. When all of a PDC stage’s cancellation resources are already busy canceling other
peaks, subsequent PDC stages will again detect and cancel the missed peak, if that stage’s resources allow.
The GC1115 contains a total of 32 cancellation pulse generators (also called cancelers) per channel that are
allocated in groups of four to PDC stages. The normal allocation of the 32 cancelers is as follows:
• PDC Stage 1 uses 4 cancelers
• PDC Stage 2 uses 8 cancelers
• PDC Stage 3 uses 12 cancelers
• PDC Stage 4 uses 8 cancelers
The allocation of cancelers should correspond to the detection threshold (DETECT_TSQD) for each PDC Stage.
A PDC stage with a lower detection threshold will find more peaks than a PDC stage with a higher detection
threshold. As mentioned above, the detection threshold for PDC stages 1 and 2 is set to detect and cancel the
largest peaks (typically those above 8 or 9 dB PAR), while PDC stages 3 and 4 detect and cancel peaks at the
desired output PAR (typically between 6 and 8 dB). Since the largest peaks are the ones that would cause the
most distortion in the PA if they were not cancelled, setting the detection threshold higher in PDC Stages 1 and
2 ensures that the large peaks will be canceled.
Cancellation pulse coefficients are stored in special GC1115 RAM blocks. The GC1115 RAM blocks can store
up to 256 real or 128 complex cancellation pulse coefficients and their associated first and second derivatives.
Cancellation coefficients are signed, 12-bit integer value between –2048 and +2047. Each cancellation
coefficient RAM supports up to four cancelers. Cancellation pulse generator RAMs assigned to the same stage
normally contain the same coefficients, but each of the GC1115’s eight cancellation RAMs may contain different
coefficients. The cancellation pulse coefficients are normalized, i.e. the largest coefficient magnitude is always
+2047 (1.0). However, both the magnitude and the phase of the normalized cancellation pulses are adjusted
before the cancellation pulse is added to the input waveform during peak cancellation, ensuring proper
alignment with the samples of the input signal.
While the subtraction of cancellation pulses reduces the peak amplitude in the region immediately surrounding a
detected peak, this subtraction may also introduce new, smaller, over-threshold peaks away from the region of
the peak. This phenomenon is called peak regrowth. Peak regrowth is an infrequent phenomenon whose effects
are mitigated by using four serial PDC stages. If Stage N causes peak regrowth, Stage N+1 and subsequent
stages will detect and cancel the regrowth peaks, assuming the PDC stages have available pulse cancelers.
Regrowth peaks, when they occur, are usually just a few percent (or less) over the PAR threshold.
Following the four PDC stages, the GC1115 also contains an interpolator and a soft limiter. When enabled, the
interpolator supports 2x real, 2x complex, and 4x real interpolation. The real output modes also modulate the
signal by f_s/4. The soft limiter acts as a fixed-length AGC that optionally attenuates the input signal over a
fixed-length window of up to 33 samples surrounding the detected over-threshold peak. Under nearly all
circumstances, the soft limiter should be bypassed.
The GC1115 operates most efficiently when there are at least 2.5 complex samples per Hertz of bandwidth. For
example, a four-carrier (20 MHz) 3G stack should be sampled at no less than 50 Msamp/sec, while a
single-carrier 3G system can be sampled as low as 12.5 Msamp/sec. Under certain circumstances, the input
signal can be represented using as few as 2.0 complex samples per Hz.
The GC1115 contains two flexible, user-programmable snapshot RAMs. The snapshot memories operate either
in capture or in histogram mode. In the capture mode, 1024 consecutive I and Q samples are stored in one of
the snapshot RAMs. During capture mode, 32 bits per complex sample (16 bits I, 16 bits Q) are captured. The
lower 2 LSBs of each real and imaginary 18-bit sample are dropped.
Each of the two snapshot RAMs receives samples from one of five user-selected test probe points:
1. At the GC1115 input
2. After PDC Stage 1
Submit Documentation Feedback
7