SDR（软件无线电）具有极大的适应性，允许随意改变模式或波形。本设计方案将焦点放在适度带宽SDR的“激励源”部分（图1）。RF载波或发射机IF加入积分调节器，根据设计细节，经调制的输出为进一步频率转化或扩大而退出。DSP部分一半用于分析基带信号——既然这样，信号分为实部和虚部。这些信号源于语音经带ADC麦克风的输出，或计算机中的数据。不管信号来源，DSP完成一串数字的计算，实现滤波，也许增加了信号音调或打包数据，转换数据串到最后I和Q的调制信号。对适度带宽，立体声Σ-ΔDAC或编码器提供转换到模拟信号，对信号实现一些额外滤波。积分调制器由一对混频器组成，所以这些滤波器通常是必要的。这些混频器将基带频率的任何噪声直接转化到调制器输出。

　　输出噪声是个问题。FCC（美国联邦通信委员会）规定一些设备上，如地面移动无线电，频谱遮罩或邻近信道功率比的需求。这些需求根据信道带宽和发射频率改变，控制允许发送的频谱。它们的功能通常一致，但在发射机邻近信道上限制与其他用户冲突。满足频谱遮罩是调制的需求；不 能保证无线电没有满足需求，且没有这个证明，是不能合法销售的。图2显示一个频谱遮罩的例子，47 CFR 90.210 G，X轴显示距信道中心的偏移量，Y轴显示未调制的载波输出。这个遮罩应用于800MHz SMRS（专用移动无线电服务），其信道分别为25kHz，而信号仅占据20kHz。

　　未调制载波首先传送到遮罩中心和顶部，与相应发射机的输出功率相适应。然后，开始调制，传播频谱。所得频谱必定降到所有位置的遮罩线位置以下。

　　图2中闭合检查显示一些有趣的特征。在载波痕迹上，采样频率的毛刺出现在距中心±19.2 kHz。被调制频谱也很有趣。Σ-ΔDAC中滤波器在大约±10kHz处引起几乎垂直下降。约±12kHz出现隆起，随频谱增加逐渐下降，这是由高功率放大器的非线性引发的。

　　许多适度带宽SDR需要在Σ-ΔDAC的单端输出和典型平衡输入积分调制器之间转化。常常需要跟随带硬件滤波器的DAC输出，其消除DAC的高频噪声，并确保满足频谱遮罩需要。更复杂的事情，DAC的最佳共模和差模输出电压很有可能与调制器需要不同。简单的缩放比例因数与共模和差模电压无关。

　　考虑所有常规方法，每个I或Q通道都需要带多滤波器的四运算放大器。滤波器需要匹配精密器件，确保载波和单边频带抑制——理想积分调制器的关键——作为基带频率不能降级。另一方面，Linear技术公司的LTC1992，用单一器件解决问题。在其数据手册中，Linear显示了完全平稳的方法（参考文献1）。

　　然而，其关闭，完全平稳的方法不是必须的。图3电路，在输出信道和消除一些危险器件匹配需求之间，具有极好的相位和幅值平衡。管脚2设置需要的共模输出电压，DAC的参考电压通过管脚8的输入电阻连接。注意到任何输入和参考电压之间输出的匹配不当，将引起不对称摆动。这个应用旁路管脚7。滤波器为有源单极电路，级连反向Sallen-Key滤波器，但是其他拓扑也是可行的。

　　图4显示对地正通道的被测频率响应。显然6dB缺口是仅着眼于一半差分输出电压的结果；当检查全平稳输出时，网络增益为0dB。图5显示理想等振幅和正负输出之间180°相移的测量方差。在临界300Hz到3kHz的范围内均小于0.1dB和0.1°。即使在50kHz，误差也小于0.5dB和1°。

　　英文原文：

　　Filter simplifies software-defined radio

　　A Linear Technology LTC1992 filter replaces four op amps in the signal chain of a software-defined radio.

　　John Wendler and Ray Tremblay, Tyco Electronics, M/A-Com Wireless Systems, Lowell, MA; Edited by Charles H Small and Fran Granville -- EDN, 12/3/2007

SDRs (software-defined radios) provide enormous flexibility, permitting you to change modes or waveforms at will. This Design Idea focuses on the “exciter” portion of a moderate-bandwidth SDR (Figure 1). The RF carrier or transmitter IF enters the quadrature modulator, and the modulated output exits for further frequency translation or amplification, depending on the details of the design. The DSP section generally works with analytic signals—in this case, signals with real and imaginary parts—at baseband. These signals may have started out as a voice speaking into a microphone that attaches to an ADC, or they may have started out as data from a computer. Regardless of the signals’ origin, the DSP performs calculations on the stream of numbers, performing filtering, perhaps adding signaling tones or packetizing the data, and converting the stream into the final I and Q modulating signals. For moderate bandwidths, a stereo sigma-delta DAC or codec provides the conversion to analog signals and performs some add
itional filtering on the signal. Such filtering is often necessary because the quadrature modulator comprises a pair of mixers. These mixers translate any noise at baseband frequencies directly to the modulator’s output.

　　Output noise is problematic. The FCC (Federal Communications Commission) sets spectral masks or adjacent-channel-power-ratio requirements on some services, such as land mobile radio. These requirements govern the allowed spectrum of a transmission and vary according to the bandwidth of the channel and the frequency of transmission. Their function is always the same, however: They limit the interference to other users on nearby channels to the transmitter. Meeting the spectral mask is a regulatory requirement; you cannot certify a radio without proving that it meets this requirement, and, without this certification, you cannot legally sell it. Figure 2 shows a sample spectral mask, 47 CFR 90.210 G, with a normalized X axis to show the offset from the center of the channel and a normalized Y axis to show the unmodulated carrier output. This mask applies to the 800-MHz SMRS (specialized-mobile-radio service) in which channels are 25 kHz apart but signals can occupy only 20 kHz.

　　The unmodulated carrier first transmits at the center of the mask, and the top of the mask adjusts to correspond with the output power of the transmitter. You then turn on the modulation, thereby spreading the spectrum. The resulting spectrum must fall below the mask line in all places.

　　A close examination of Figure 2 shows some interesting features. On the carrier trace, the sampling-frequency spurs appear at ±19.2 kHz away from the center. The modulated spectrum is also interesting. The filter in the sigma-delta DAC causes the nearly vertical roll-off at approximately ±10 kHz. The mounds that appear around ±12 kHz and gradually roll off are spectral regrowth, which nonlinearities in the high-power amplifier cause.

　　Many moderate-bandwidth SDRs need a translator between the sigma-delta DAC’s single-ended output and a typical balanced-input quadrature modulator. It is frequently desirable to follow up the DAC output with a hardware filter that removes the DAC’s high-frequency noise and ensures compliance with spectral-mask requirements. Further complicating things, the optimal common- and differential-mode output voltages of the DAC are likely to differ from those that the modulator requires. An easy scaling factor does not relate common- and differential-mode voltages.

　　Handling all of these considerations with a conventional approach can require as many as four operational amplifiers with multiple filter sections per I or Q channel. The filters require close component matching to guarantee that carrier and single-sideband suppression—key measures of quadrature-modulator ideality—do not degrade as a function of baseband frequency. The Linear Technology LTC1992, on the other hand, addresses the problem in a single section. Linear shows a fully balanced approach to the problem in its data sheet (Reference 1).

　　It turns out, however, that a fully balanced approach is unnecessary. The circuit in Figure 3 has excellent phase and amplitude balance between the output channels and eliminates some critical component-matching requirements. Pin 2 is set for the desired common-mode output voltage, and the DAC’s midpoint voltage connects through an input resistor to Pin 8. Note that any mismatch between the input voltage and the midpoint voltage appears at the outputs and causes asymmetrical swing. This application bypasses Pin 7. The filter is a passive single-pole circuit cascaded with an inverting Sallen-Key filter, but other topologies are feasible.

　　Figure 4 shows the measured frequency response of the positive channel with respect to ground. The apparent 6-dB loss is a result of looking at only half the differential-output voltage; when you examine the full balanced output, the net gain is 0 dB. Figure 5 shows the measured deviation from an ideal equal-amplitude, 180° phase shift between the positive and the negative outputs. The agreement in the critical 300-Hz to 3-kHz range is less than 0.1 dB and 0.1°. Even at 50 kHz, the error is less than 0.5 dB and 1°.

　　Acknowledgment

　　The authors gratefully acknowledge the assembly assistance of Deb Girard.

　　Reference

　　“LTC1992: Fully Differential Input/Output Amplifier/ Driver,” Linear Technology, July 2003.