GC4116 Datasheet, PDF(15/57 Page) List of Unclassifed Manufacturers

English

English German Russian Spanish Italian Polish Chinese Japanese Korean French Portuguese	Language :

GC4116 Datasheet, PDF (15/57 Pages) List of Unclassifed Manufacturers – MULTI-STANDARD QUAD DUC CHIP

◁

GC4116 MULTI-STANDARD QUAD DUC CHIP

DATA SHEET REV 1.0

3.5 THE SUM TREE

As shown in Figure 1, the mixer outputs are rounded to

20 bits and put into the LSBs of a 23 bit sum tree. The sum

tree adds all four up-convert channels together. The 23 bit

sum tree output is shifted down by four bits and rounded to

19 bits before being added into the LSBs of the external 22

bit sum input. The final 23 bit sum is either saturated to 22

bits (the MSB is checked for overflow) and output from the

chip as 22 bits, or is scaled up by 0 to 7bits, rounded into the

8, 10, 12, 14, 16, 18, 20, or 22 MSBs and then output from

the chip.

The sum tree gain is equal to 2SUM_SCALE-7, where

SUM_SCALE is 0 to 7 (See address 19 of the IO control

page). Overflows in the sum tree are saturated to plus or

minus full scale.

The latency from SUMI[0:21] to SUMO[0:21] is eight

clock cycles.

3.6 OVERALL GAIN

The overall gain of the chip is a function of the input gain

setting (G), the sum of the programmable filter coefficients

(PFIR_SUM described in Section 3.3.1), the amount of

interpolation in the CIC filters (N described in Section 3.3.3),

the scale circuit settings in the CIC filter (SCALE and

BIG_SHIFT described in Section 3.3.3), and the sum tree

scale factor (SUM_SCALE described in Section 3.5). The

overall gain, excluding any resampler gain, is:

= { }{ } GAIN

ï£±

ï£²

ï£³

1--G-2---8--

ï£¼ï£±

ï£½ï£²

ï£¾ï£³

P----F---I6--R-5---_5---S3---6U-----M----

ï£¼

ï£½

ï£¾

N4 2â(SCALE + 12 Ã BIG_SHIFT + 3)

2SUM_SCALE-7

where G and PFIR_SUM can be different for each channel,

but N, SCALE, BIG_SHIFT, and SUM_SCALE are common

to all channels. The resamplers gain, which precedes the

input gain, is discussed in Section 3.7.6.

The optimal gain setting is one which will keep the

amplitude of the data within the channel as high as possible

without causing overflow. For random amplitude data the

recommended gain target is to keep the root-mean-squared

amplitude of the data close to one-fifth (0.2) full scale (a 14

dB crest factor). This level should be maintained throughout

the channel computations. This means that the products

ï£³ï£²ï£± 3--R-2---M7---6--S-8--

ï£¼

ï£½

ï£¾

ï£³ï£²ï£± 1--G-2---8--

ï£¼

ï£½

ï£¾

ï£³ï£²ï£± P----F---I6--R-5---_5---S3---6U-----M----

ï£¼

ï£½

ï£¾

and

ï£±

ï£²

ï£³

--R----M-----S---

32768

ï£¼ï£±

ï£½ï£²

ï£¾ï£³

--G------

128

ï£¼ï£±

ï£½ï£²

ï£¾ï£³

P----F---I--R----_---S---U-----M----

65536

ï£¼

ï£½

ï£¾

{

2â(

SCALE

BIG_SHIFT

}

should both be less than or equal to 0.2, where âRMSâ is the

root-mean-squared level of the input data. Other crest

factors can be used depending upon the application. For

example, a crest factor of 12 dB is adequate if the final

number of bits going to a DAC is 12 bits. In most cases the

input data will already have the correct crest factor for the

application,

in which case the ratio

ï£±

ï£²

ï£³

3--R-2---M7---6--S-8--

ï£¼

ï£½

ï£¾

will be

equal

to the

crest factor (e.g., 0.2) and the gain settings in the channel

should be set to unity.

In some applications the input amplitude is far from

random. For example, QPSK data has constant amplitude. In

such cases the largest gain that guarantees no overflow can

be calculated from the PFIR coefficients and normally allows

a substantially higher gain than the optimal gain for random

data of similar power.

Note that the resamplerâs gain can be used to increase

or decrease the RMS input level to the channels.

The sum tree adds the four channels within a single

GC4116 together and then adds in sums from other chips

using the sum I/O ports. The 22 bit sum I/O path guarantees

that no overflow will occur for systems with 8 chips (32

channels) or less. The final chip in the chain should then shift

and round the result to optimize the performance of the D/A.

Since this represents the sum of many channels the gain

should be set with a 14 dB crest factor.

The 14 dB crest factor assumes that the channels can

be treated as uncorrelated signals which will result in a

random, uniform amplitude distribution. If M signals are

correlated, however, the amplitude gain can be M and the

sum tree gain should be set to --1-- . Examples of correlated

signals are pure tones or modem signals that have been

synchronized so that they might peak at the same time.

These signals, however, require a much smaller crest factor,

such as 3 dB for pure tones and 6 dB for modem signals. In

this case the crest factor of 14 dB will absorb much of the

difference in gain between M and M .

If overflow does occur, then the samples are saturated

to plus or minus full scale. Overflow can be monitored using

the status register (address 14).

The values of N and BIG_SHIFT must also satisfy

2(12*BIG_SHIFT+18) â¥ N4 (see Section 3.3.3 for details). If N

and BIG_SHIFT do not satisfy this relationship, then an

overflow may occur which may not be detected.

If the auto flush mode is used, then the gain in the CIC

must be less than or equal to unity. This means that the

values of N, SCALE and BIG_SHIFT must satisfy

2(SCALE+12*BIG_SHIFT+3) â¥ N4 (see Section 3.3.3 for details).

- 10 -

APRIL 27, 2001

This document contains preliminary information which may be changed at any time without notice

▷