JP3656739B2 - Low IF receiver - Google Patents

Description

表現する。
【００２１】
【数２】

を得る。ＱＭＦフィルタバンクは
Ｈ₁(ｚ) ＝ Ｈ₀(−ｚ), (2.2)
Ｆ₀(ｚ) ＝ Ｈ₀(ｚ), (2.3)
Ｆ₁(ｚ) ＝ −Ｈ₁(ｚ) (2.4)
のとき、エイリアシング回避条件
Ｈ₀(ｚ)Ｆ₀(ｚ)＋Ｈ₁(ｚ)Ｆ₁(ｚ) ＝ ０ (2.5)
を満たし、オールパス条件
Ｈ₀(−ｚ)Ｆ₀(z)＋Ｈ₁(−ｚ)Ｆ₁(ｚ) ＝ ２ｚ^-L (2.6)
を近似的に満たす。ここでＬはフィルタの次数である。
【００２２】
図２は、実数係数のＱＭＦフィルタバンクの周波数応答を示す図である。受信信号のＤＣ成分及びイメージ周波数信号を抑圧するためには、実数係数を次のような変換式により複素係数に変換する。
ｈ₀(ｎ) ＝ exp(ｊｎφπ)ｈ_r(ｎ), (2.7)
ｈ₁(ｎ) ＝ exp(ｊｎ(φ±１)π)ｈ_r(ｎ). (2.8)
ここでφはフィルタのインパルス応答の周波数シフト量である。
【００２３】
図３は、π／４周波数シフトした時のＱＭＦフィルタバンクの周波数応答を示す図である。このように周波数シフトすることで、ＤＣ成分及びイメージ周波数信号を抑圧して、希望周波数帯域を均等に分割することができる。
【００２４】
普通、このような帯域分割フィルタはスイッチドキャパシタにより構成される。スイッチドキャパシタフィルタの特性は、寄生容量等により誤差を含む。
【００２５】
図４は、２分割の場合の参考として示す低ＩＦ方式受信機の構成を示す図である。すなわち、ここではＭ＝２，Ｎ＝２とする。アンテナ３１、ＢＰＦ３２、ＬＮＡ３３、ミキサ３４−０、３４−１、局部発振器３５、帯域分割フィルタ３７−０、３７−１、増幅器３８−０、３８−１、デシメータ３９−０、３９−１、Ａ／Ｄ変換器４０−０、４０−１、アップサンプラ４１−０、４１−１、帯域合成フィルタ４２−０、４２−１、及び復調器４７はそれぞれ図１に示されるアンテナ１１、ＢＰＦ１２、ＬＮＡ１３、ミキサ１４、局部発振器１５、帯域分割フィルタ１６−０、１６−１、…、１６−Ｍ−１、増幅器１７−０、１７−１、…、１７−Ｍ−１、デシメータ１８−０、１８−１、…、１８−Ｍ−１、Ａ／Ｄ変換器１９−０、１９−１、…、１９−Ｍ−１、アップサンプラ２０−０、２０−１、…、２０−Ｍ−１、帯域合成フィルタ２１−０、２１−１、…、２１−Ｍ−１、及び復調器２２に相当する。ここでは、希望信号帯域Ｈ₁より誤差推定４４で誤差を推定し、ＤＣオフセット通過帯域Ｈ₀の平均４３とからフィルタ係数を推定して、加算器４６から希望信号を出力する。
【００２６】
まずは、複素係数の誤差を推定する。推定はミキサ３４−０、３４−１の入力端を抵抗を介して接地し、アンテナ３１から信号が入力しない状態で行う。このとき局部発振器３５からのローカル信号をミキサ３４−０、３４−１に入力するとＤＣオフセットが発生する。これを利用して複素係数の誤差を推定する。
【００２７】
帯域分割フィルタ３７−０、３７−１にてサンプルされた既知の複素ベースバンド信号をbarｒ(ｎ)とすると
【００２８】
【数３】

ここで*は複素共役を示し、ｍは分割する帯域を示すインデックスである。またadc｛｝はＡ／Ｄ変換を示す。hatｙ_m(ｎ)をｎ＝ｎ，…，ｎ＋Ｌ−１まで求める。
【００２９】
【数４】

よりフィルタの係数hatＨmを推定し、メモリ２２ａに記憶しておく。
【００３０】
誤差を含んだ帯域分割フィルタの係数をtldｈm(ｎ)とする。tldｈo(ｎ)はＤＣオフセットが通過し、tldｈ₁(ｎ)は希望信号が通過するとする。帯域分割フィルタ３７−０、３７−１にてサンプルされた受信信号をbarｒ(ｎ)とすると
【００３１】
【数５】

ここでδｈoはフィルタｈoの誤差を示す。tldｙo(ｎ)はダウンサンプルされＡ／Ｄ変換されたのちアップサンプルされる。
【００３２】
【数６】

ここで受信信号ｒ(ｎ)が大きなＤＣオフセットを含むと仮定するとｙ_Ｕ０の推定値は
【００３３】
【数７】

式（3.19）よりＤＣオフセットを推定する。
【００３４】
【数８】

tldｙ₁(ｎ)はダウンサンプルされＡ／Ｄ変換されたのちアップサンプルされる。
【数９】

ＤＣオフセットによる干渉成分は以下のように除去する。
【００３５】
【数１０】

希望信号は以下のようにして取り出す。
【００３６】
【数１１】

図５は、フィルタの周波数シフト量とＢＥＲ（ビット誤り率）の関係を示す図である。「ＡＤＣ」は受信信号をそのままＡ／Ｄ変換した後に複素係数サブバンドフィルタによりＤＣオフセット成分を除去した場合である。この場合には他の方式よりも２倍のＡ／Ｄ変換速度が必要になる。「補償なし」は複素係数アナログサブバンドフィルタによりＤＣオフセットを除去するが誤差の補償を行わない場合である。「発明」は提案する誤差補償方式を用いた場合である。いずれの場合にも−４π／１６〜−６π／１６［rad／サンプル］あたりでＢＥＲが減少している。−６π／１６［rad／サンプル］以上にシフトするとＤＣオフセットがフィルタ後の信号に混入し、ＢＥＲが高くなる。−３π／１６［rad／サンプル］以下にシフトするとイメージ周波数信号がフィルタ後の信号に混入するためやはりＢＥＲが高くなる。
【００３７】
図６は、ＢＥＲとＥb／Ｎoの関係を示す図である。Ｅb／Ｎoとは、「ビットエネルギー（電力）対雑音電力密度」であり、１情報ビットあたりの信号対雑音電力比（Ｓ／ＮまたはＳＮＲ：Signal to Noise power Ratio）である。「ＤＣオフセットなし」はＤＣオフセットのない場合であり、ＱＰＳＫの理論値を示す。「補償なし」と「発明」を比較することにより、提案する誤差補償法を用いるとＢＥＲが大幅に改善することがわかる。ただし「ＡＤＣ」に比べてＢＥＲ＝１０^-2において約１ｄＢほど特性が劣化する。これはデシメーションをした際にＤＣオフセットのエイリアシング成分がフィルタ後の出力に混入するためである。
【００３８】
図７は、ＢＥＲとＡ／Ｄ変換器の解像度の関係を示す図である。図７よりＡ／Ｄ変換器の解像度が１０ビット以下の時には、あらかじめ複素係数サブバンドフィルタによりＤＣオフセットを抑圧したのちにＡ／Ｄ変換したほうがＢＥＲが低い。これに対し解像度が１２ビット以上ではＡ／Ｄ変換器でディジタル信号に変換したのち、フィルタ係数誤差の少ないディジタル信号処理によってＤＣオフセットを取り除いたほうがＢＥＲが少ないことを示している。
【００３９】
図８は、本発明の実施の形態による低ＩＦ方式受信機の構成を示す図である。本実施の形態の低ＩＦ方式受信機は、アンテナ７１、ＢＰＦ７２、ＬＮＡ７３、ミキサ７４−１、７４−１、局部発振器７５、遅延器７６−０、７６−１、…、７６−Ｎ−２、デシメータ７７−０、７７−１、…、７７−Ｎ−１、アナリシスフィルタバンク７８、増幅器７９−０、７９−１、…、７９−Ｎ−１、Ａ／Ｄ変換器８０−０、８０−１、…、８０−Ｎ−１、及びメモリ８１ａを有するデコリレータ８１から成る。
【００４０】
アンテナ７１を介して受信された信号はＢＰＦ７２を通りＬＮＡ７３でＲＦ増幅されミキサ７４により直角位相関係にある２つの低ＩＦ信号に変換される。変換された信号は遅延器７６−０、７６−１、…、７６−Ｎ−２により順次遅延された後に、デシメータ７７−０、７７−１、…、７７−Ｎ−１により１／Ｎにダウンサンプルされて、アナリシスフィルタバンク７８に入力される。これによりＡ／Ｄ変換器８０−０、８０−１、…、８０−Ｎ−１の変換速度を１／Ｎにする。アナリシスフィルタバンク７８は入力信号を複数の周波数成分に分割して各サブバンドの信号を出力する。各サブバンドの信号は増幅器７９−０、７９−１、…、７９−Ｎ−１で増幅され、Ａ／Ｄ変換器８０−０、８０−１、…、８０−Ｎ−１でディジタル信号に変換されて、デコリレータ８１で入力信号から相互干渉成分を取り除く。
【００４１】
ここで、送信信号の周波数成分がｄ＝｛ｄ₀,ｄ₁,…,ｄ_N-1｝^Tを持つと仮定する。このとき受信信号は以下のように表される。
【００４２】
【数１２】

ここでΔＴはサンプリング間隔、Ｎはアナリシスフィルタバンク７８の入力数である。アナリシスフィルタバンク７８の係数に誤差がない場合にはそのｋ番目の出力は以下のようになる。
【００４３】
【数１３】

ただし増幅器７９−０、７９−１、…、７９−Ｎ−１の増幅度は簡単化のため１とした。
【００４４】
式（２）においてはアナリシスフィルタの係数誤差及び雑音の影響を無視している。実際にはアナリシスフィルタの係数は誤差を含み、また受信信号には雑音が含まれる。
【００４５】
【数１４】

ここでｃ_nkはｋ番目の帯域を抽出するフィルタのｎ番目の係数の誤差を示し、ηは熱雑音を表す。式（３）を行列表示すると
【００４６】
【数１５】

さらにデシメーションされた信号が各Ａ／Ｄ変換器８０−０、８０−１、…、８０−Ｎ−１を通過する時には量子化雑音が加わる。これをζ＝｛ζ₀,ζ₁,…,ζ_N-1｝^Tとすると
【００４７】
【数１６】

本実施の形態においては、予めＤＣオフセットを利用してアナリシスフィルタバンク７８の係数を推定して、設計値と推定値を比較し、各周波数分割した信号間の相互相関行列を計算し、相互相関行列の逆行列を計算して、メモリ８１ａに記憶しておく。デコリレータ８１では、hatＲの逆行列をＡ／Ｄ変換器８０−０、８０−１、…、８０−Ｎ−１の出力に掛け算する。すなわち
【００４８】
【数１７】

デコリレータの出力は各周波数信号成分ｄ_k、及び熱雑音ηと量子化雑音ζが相互相関の影響を受けた信号を含む。
【００４９】
ここで、デコリレータを用いない従来型受信機の誤り率特性を計算する。式（１０）より受信した希望信号と他の周波数帯域からの干渉成分は以下のように表される。
【００５０】
【数１８】

ここでRe｛ａ｝は複素数ａの実部を表し、［Ａ］_kはＮ×１行列Ａのｋ番目の要素を意味する。また熱雑音の影響による雑音の分散は式（８）より
【００５１】
【数１９】

ここでσ²は熱雑音の分散である。したがって
【００５２】
【数２０】

またＡ／Ｄ変換器の入力最大値の絶対値をＡ_Mとし、解像度をＢqビットとし最初の１ビットは正負の符号に用いるとすると量子化雑音ζ_kは［−(Ａ_M／２^Bq),(Ａ_M／２^Bq)］の間で一様分布する。
【００５３】
ｄ_kをＢＰＳＫ（Binary Phase Shift Keying）信号と仮定すると条件付誤り率は以下のように求められる。
【００５４】
【数２１】

ここで［Ａ］k,kはＮ×Ｎ行列Ａの（ｋ,ｋ）番目の要素を意味する。
【００５５】
【数２２】

と変数変換し、誤差関数を展開すると
【００５６】
【数２３】

ここで
【００５７】
【数２４】

したがって補償を行った場合の受信機の誤り率は
【００５８】
【数２５】

ここで、本実施の形態のデコリレータ型誤差補償法の特性を解析する。式（１０）において雑音成分の分散を計算する。熱雑音については
【００５９】
【数２６】

また量子化誤差については各Ａ／Ｄ変換器で発生する量子化誤差の分散が等しくλ²であると仮定すると
【００６０】
【数２７】

で与えられる。量子化雑音は一様分布であるが、デコリレータの係数が掛け合わされることによりガウス雑音とモデル化される。したがってｋ番目の周波数帯域におけるＳＮＲ（ＳＮ比）は以下のように求められる。
【００６１】
【数２８】

ここで［Ａ］kkはＮ×Ｎ行列Ａの（ｋ,ｋ）番目の要素を意味する。仮にｄ_kがＢＰＳＫ変調された信号であり、雑音成分がガウス性だと仮定するとＢＥＲは以下のように求まる。
【００６２】
【数２９】

ここで、シミュレーションとして、アナリシスフィルタバンク７８の各係数に一様分布に従って発生する［−０.０１,０.０１］の誤差を付加した場合の、１００００回の平均誤り率を計算する。
【００６３】
希望信号はＢＰＳＫ変調して送信され、受信されたのちＩＦ＝３π／８［rad／サンプル］にダウンコンバージョンされるとする。通信路は白色ガウス雑音路を仮定する。コヒーレント受信を仮定し、Ａ／Ｄ変換器の与えられた解像度のうち１ビットは正負の符号を表すのに用いられ、Ａ_M＝２hatｄ_k又はＡ_M＝２ｓ_k(ｄ)とする。イメージ周波数信号はＢＰＳＫ信号が−３π／８［rad／サンプル］に受信されるとする。イメージ周波数信号の電力は希望信号に対して６０［dB］大きいとする。
【００６４】
図９は、本実施の形態においてＢＥＲとＥb／Ｎoの関係を示す図である。ここでは、Ａ／Ｄ変換器の解像度を１０［bit］とした。図よりデコリレータを用いない従来型受信機の場合、フィルタの係数誤差によりＢＥＲ特性がフロアを示し、Ｅb／Ｎoの増加に対してほとんど変化しない。これはイメージ周波数信号が希望信号に対して係数誤差のために干渉するためである。これに対してデコリレータ型補償法を用いた場合にはＢＥＲ特性がほぼＢＰＳＫの理論値を示している。これはデコリレータがイメージ周波数信号の干渉を取り除くためである。
【００６５】
なお、本発明は上記実施の形態に限定されるものではない。
【００６６】
上述の実施の形態では、ミキサの入力を抵抗を介して接地することでローカル信号を利用して既知信号を作っているが、既知の信号を発生する局部発振器を別途設けて供給するようにしても良い。
【００６７】
また、フィルタの構成をフィルタバンクとするのではなく、ウェーブレットを用いてもよい。すなわち、一度に多数に帯域分割するのではなく、例えば２分割ずつ繰り返し分割していくように、階層的に分割するようにしても良い。こうすることにより例えば分割する所定の階層を短絡するスイッチング等により分割の細かさを切換えることができる。
【００６８】
【発明の効果】
以上のように、本発明によれば、アナログ部である帯域分割フィルタの係数誤差をディジタル信号処理によって補償することができるとともに、特に、アナログフィルタの誤差により直交する周波数信号間で相互に引き起こす干渉をも除去することができる。これにより、Ａ／Ｄ変換器の変換速度及び解像度に関する要求を緩和し、広帯域通信システムに対応することができる。さらに、並列信号処理を可能にし、低消費電力の受信構成を実現できる。
【図面の簡単な説明】
【図１】 参考として示す低ＩＦ方式受信機の構成を示す図である。
【図２】 実数係数のＱＭＦフィルタバンクの周波数応答を示す図である。
【図３】 π／４周波数シフトした時のＱＭＦフィルタバンクの周波数応答を示す図である。
【図４】 ２分割の場合の参考として示す低ＩＦ方式受信機の構成を示す図である。
【図５】 フィルタの周波数シフト量とＢＥＲの関係を示す図である。
【図６】 ＢＥＲとＥb／Ｎoの関係を示す図である。
【図７】 ＢＥＲとＡ／Ｄ変換器の解像度の関係を示す図である。
【図８】 本発明の実施の形態による低ＩＦ方式受信機の構成を示す図である。
【図９】 本実施の形態においてＢＥＲとＥb／Ｎoの関係を示す図である。
【図１０】 従来のダイレクトコンバージョン受信機の構成を示す図である。
【図１１】 従来の低ＩＦ方式受信機の構成を示す図である。
【符号の説明】
１１、３１、７１ アンテナ
１７、３８、７９ 増幅器
１８、３９、７７ デシメータ
２０、４１ アップサンプラ
２１、４２ 帯域合成フィルタ[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a low IF receiver, and more particularly to a low IF receiver that uses a complex coefficient filter bank to suppress DC offset and image frequency signals.
[0002]
[Prior art]
In recent years, with the trend toward broadband networks, there has been an increasing demand for broadband communication systems. In order to realize a multi-mode / multi-band terminal, a direct conversion receiver has been studied.
[0003]
FIG. 10 is a diagram showing a configuration of a conventional direct conversion receiver. In the direct conversion method, the intermediate frequency (IF) stage and the like are deleted, and the signal received via the antenna 51 is band-amplified by the amplifier filter 52 and mixed with the local oscillation signal from the local oscillator 54 by the mixer 53 to directly base the signal. It is converted into a band signal. The converted signal passes through the amplifier filter 55 and is input to the A / D converter 56. The output of the A / D converter 56 is digital signal processed and demodulated. Since this method does not use an IF filter, it is easy to increase the bandwidth of the receiver, increasing the flexibility of the receiver and simultaneously enabling one chip. However, in the direct conversion method, since the reception frequency and the local frequency are equal, the problem of DC offset has been pointed out. The DC offset is a problem in which a part of the local signal leaks from the input side of the mixer, returns to the mixer again, and self-mixes with the local signal to cause an offset in the DC component. Alternatively, a part of the local signal is emitted to the outside through the antenna, and the reflected wave is received and mixed with the local signal.
[0004]
Low IF methods have been proposed to alleviate the DC offset problem (J. Crols and MSJ Steyaert, "Low-IF Topologies for High-Performance Analog Front Ends of Fully Intefrated Receivers," IEEE Trans. On Circuits and Systems, vol.45, no.3, pp.269-282, March 1998.).
[0005]
FIG. 11 is a diagram illustrating a configuration of a conventional low-IF receiver. A signal received via the antenna 61 is mixed with a local oscillation signal from a local oscillator 65 by a mixer 64 through a band pass filter (BPF) 62 and an LNA 63, and converted into a low IF signal. The converted signal passes through the IF filter 66 and the amplifier 67 and is input to the A / D converter 68. The output of the A / D converter 68 is digital signal processed (DSP) by the demodulator 69 and demodulated. In the low IF method, since the desired signal does not exist near DC, the DC offset does not interfere with the desired signal. However, in order to A / D convert and remove the DC offset and the image frequency signal as they are, it is necessary to convert the DC offset and the image frequency signal and the desired signal into a digital signal at the same time. Therefore, an A / D converter having a high conversion speed is required. Required.
[0006]
The improvement trend of the A / D converter is about 1.5 bits in 8 years, and considering the tendency to decrease the resolution by 1 bit every time the conversion speed is doubled, the conversion is equivalent to the improvement degree of the signal processing speed. There is no improvement in speed.
[0007]
As a method for improving the conversion speed of an A / D converter, a method of dividing a signal band using a filter bank has been proposed (R. Khoini-Poorfard, LB Lim, and DA Johns, "Time-Interleaved Oversampling A / D Converters: Theory and Practice, "IEEE Trans. On Circuits and Systems-II: Analog and Digital Signal Processing, vol.44, no.8, pp.634-645, August 1997.). By dividing the band and performing A / D conversion in parallel, the conversion speed required for each A / D converter can be reduced. At the same time, it is possible to relax the resolution requirement of the A / D converter by applying the optimized AGC to each subband signal.
[0008]
[Problems to be solved by the invention]
However, in this A / D conversion method, each complex coefficient sub-band filter is constituted by a switched capacitor filter (SCF) or the like, and an error occurs in the coefficient of the filter, and the characteristics in the stop band are deteriorated. Then, the DC offset and the image frequency signal may interfere with the desired signal.
[0009]
In view of the above problems, the present invention configures a low IF receiver using a band division A / D converter, and reduces the interference component that has passed through each subband filter. An object is to provide an IF receiver.
[0010]
[Means for Solving the Problems]
The low IF receiver according to the present invention divides a low IF signal into N (N is an integer of 2 or more) band signals, and performs analog / digital conversion on the output of the band division analog filter. An A / D converter, known signal supply means for supplying a low IF known signal to the low IF stage, and when the known signal is supplied to the band division analog filter by the known signal supply means ,
hat d = hat Rd
Here, hat d: an (N , 1) matrix having the output signal amplitude of each band as a component, and d: an (N , 1) matrix having the input signal amplitude of each band as a component
A memory that stores hat R ⁻¹ , which is an inverse matrix to the (N 1 , N) matrix hat R , and a decorrelator that reads the contents of the memory and multiplies the output of the A / D converter by hat R ⁻¹ ; A digital demodulator for demodulating the decorrelator output.
[0011]
Further, the known signal supply means uses a local oscillator for converting the received signal frequency to a low IF, so that the local oscillator can be shared for receiving frequency conversion and for storing filter characteristics.
[0012]
Further, the band division analog filter has a filter bank configuration, and can perform high-speed processing.
[0013]
Further, the band division analog filter has a wavelet configuration, so that it can flexibly cope with a change in filter division characteristics.
[0014]
In addition, since the band division analog filter has a complex coefficient filter configuration, the band of the division filter can be effectively used by suppressing the DC component and the image frequency.
[0016]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.
[0017]
FIG. 1 is a diagram illustrating a configuration of a low-IF receiver shown as a reference . This low IF receiver includes an antenna 11, a BPF 12, an LNA (low noise amplifier) 13, a mixer 14, a local oscillator 15, band division filters 16-0, 16-1, ..., 16-M-1, an amplifier 17- , 17-M-1, decimators 18-0, 18-1, ..., 18-M-1, A / D converters 19-0, 19-1, ..., 19-M- 1, up-samplers 20-0, 20-1, ..., 20-M-1, band synthesis filters 21-0, 21-1, ..., 21-M-1, and a demodulator 22 having a memory 22a.
[0018]
A signal received via the antenna 11 passes through the BPF 12 and is converted into a low IF signal by the mixer 14. The converted signals are input to complex coefficient band division filters 16-0, 16-1,..., 16-M-1 from H ₀ to H _M−1 . Band division filters 16-0, 16-1, ..., 16-M-1 separate the desired signal from the DC offset and image frequency signals. The input signal is down-sampled to 1 / N by decimators 18-0, 18-1, ..., 18-M-1. As a result, the conversion speed of the A / D converters 19-0, 19-1, ..., 19-M-1 is set to 1 / N. Further, the resolution of the A / D converters 19-0, 19-1,..., 19-M-1 can be changed according to the signals received in each subband. The samples of the signals converted by the A / D converters 19-0, 19-1,..., 19-M-1 are N samples by the upsamplers 20-0, 20-1,. The signals are converted into samples, and the signals of the respective bands are synthesized by the complex coefficient band synthesis filters 21-0, 21-1,..., 21-M-1, and demodulated by the demodulator 22.
[0019]
Many proposals such as a DFT (Digital Fourier Transform) filter bank and a CQF (Conjugate Quadrature Filter) filter bank have been made as filter banks. In particular, when M = 2, a QMF (Quadrature Mirror Filter) bank is known. When z conversion of the input signal x (n) and the output signal hatx (n) is X (z) and hatX (z),
In the text below,
[0020]
[Expression 1]

Express.
[0021]
[Expression 2]

Get. QMF filter bank is H ₁ (z) = H ₀ (−z), (2.2)
F ₀ (z) = H ₀ (z), (2.3)
F ₁ (z) = − H ₁ (z) (2.4)
The aliasing avoidance condition H ₀ (z) F ₀ (z) + H ₁ (z) F ₁ (z) = 0 (2.5)
The all-pass condition H ₀ (−z) F ₀ (z) + H ₁ (−z) F ₁ (z) = 2z ^−L (2.6)
Is approximately satisfied. Here, L is the order of the filter.
[0022]
FIG. 2 is a diagram illustrating the frequency response of a real number coefficient QMF filter bank. In order to suppress the DC component of the received signal and the image frequency signal, the real coefficient is converted into a complex coefficient by the following conversion formula.
h ₀ (n) = exp (jnφπ) h _r (n), (2.7)
h ₁ (n) = exp (jn (φ ± 1) π) h _r (n). (2.8)
Here, φ is the frequency shift amount of the impulse response of the filter.
[0023]
FIG. 3 is a diagram showing the frequency response of the QMF filter bank when the frequency is shifted by π / 4. By shifting the frequency in this way, it is possible to suppress the DC component and the image frequency signal and divide the desired frequency band evenly.
[0024]
Usually, such a band division filter is constituted by a switched capacitor. The characteristics of the switched capacitor filter include errors due to parasitic capacitance and the like.
[0025]
FIG. 4 is a diagram showing a configuration of a low IF receiver shown as a reference in the case of two divisions. That is, here, M = 2 and N = 2. Antenna 31, BPF 32, LNA 33, mixers 34-0 and 34-1, local oscillator 35, band division filters 37-0 and 37-1, amplifiers 38-0 and 38-1, decimators 39-0 and 39-1, A / D converters 40-0 and 40-1, upsamplers 41-0 and 41-1, band synthesis filters 42-0 and 42-1, and demodulator 47 are antenna 11, BPF 12, and LNA 13 shown in FIG. , Mixer 14, local oscillator 15, band division filters 16-0, 16-1, ..., 16-M-1, amplifiers 17-0, 17-1, ..., 17-M-1, decimators 18-0, 18 -1, ..., 18-M-1, A / D converters 19-0, 19-1, ..., 19-M-1, upsamplers 20-0, 20-1, ..., 20-M-1, Band synthesis filters 21-0, 21-1,... 1-M-1, and corresponds to the demodulator 22. Here, the error is estimated by the error estimation 44 from the desired signal band H _{1, the} filter coefficient is estimated from the average 43 of the DC offset pass band H ₀ , and the desired signal is output from the adder 46.
[0026]
First, the error of the complex coefficient is estimated. The estimation is performed in a state where the input terminal of the mixers 34-0 and 34-1 is grounded via a resistor and no signal is input from the antenna 31. At this time, when a local signal from the local oscillator 35 is input to the mixers 34-0 and 34-1, a DC offset is generated. Using this, the error of the complex coefficient is estimated.
[0027]
Assuming that a known complex baseband signal sampled by the band division filters 37-0 and 37-1 is barr (n),
[Equation 3]

Here, * indicates a complex conjugate, and m is an index indicating a band to be divided. Adc {} indicates A / D conversion. Haty _m to (n) n = n, ... , determined to n + L-1.
[0029]
[Expression 4]

Further, the filter coefficient hatHm is estimated and stored in the memory 22a.
[0030]
The coefficient of the band division filter including the error is assumed to be tldhm (n). It is assumed that a DC offset passes through tldho (n) and a desired signal passes through tldh ₁ (n). When the reception signal sampled by the band division filters 37-0 and 37-1 is barr (n),
[Equation 5]

Here, δho indicates an error of the filter ho. tldyo (n) is downsampled and A / D converted and then upsampled.
[0032]
[Formula 6]

Assuming that the received signal r (n) includes a large DC offset, the estimated value of _YU is
[Expression 7]

The DC offset is estimated from equation (3.19).
[0034]
[Equation 8]

tldy ₁ (n) is downsampled, A / D converted, and then upsampled.
[Equation 9]

The interference component due to the DC offset is removed as follows.
[0035]
[Expression 10]

The desired signal is extracted as follows.
[0036]
[Expression 11]

FIG. 5 is a diagram showing the relationship between the frequency shift amount of the filter and the BER (bit error rate). “ADC” is a case where a DC offset component is removed by a complex coefficient subband filter after A / D converting the received signal as it is. In this case, an A / D conversion speed twice that of other methods is required. “No compensation” is a case where DC offset is removed by a complex coefficient analog subband filter, but error compensation is not performed. The “invention” is a case where the proposed error compensation method is used. In either case, the BER is reduced around −4π / 16 to −6π / 16 [rad / sample]. When shifting to −6π / 16 [rad / sample] or more, the DC offset is mixed in the signal after filtering, and the BER is increased. When shifting to −3π / 16 [rad / sample] or less, the image frequency signal is mixed in the signal after filtering, so that the BER is also increased.
[0037]
FIG. 6 is a diagram showing the relationship between BER and Eb / No. Eb / No is “bit energy (power) vs. noise power density”, which is a signal to noise power ratio (S / N or SNR) per information bit. “No DC offset” is a case where there is no DC offset, and indicates a theoretical value of QPSK. By comparing "no compensation" with "invention", it can be seen that the BER is greatly improved using the proposed error compensation method. However, the characteristic deteriorates by about 1 dB at BER = 10 ^{−2 as} compared with “ADC”. This is because the aliasing component of the DC offset is mixed into the output after filtering when decimation is performed.
[0038]
FIG. 7 is a diagram showing the relationship between the BER and the resolution of the A / D converter. From FIG. 7, when the resolution of the A / D converter is 10 bits or less, the BER is lower when the A / D conversion is performed after the DC offset is previously suppressed by the complex coefficient subband filter. On the other hand, when the resolution is 12 bits or more, the BER is smaller when the digital offset is removed by digital signal processing with a small filter coefficient error after the digital signal is converted by the A / D converter.
[0039]
FIG. 8 is a diagram showing a configuration of a low IF receiver according to the embodiment of the present invention. The low IF receiver of the present embodiment includes an antenna 71, a BPF 72, an LNA 73, mixers 74-1, 74-1, a local oscillator 75, delay units 76-0, 76-1, ..., 76-N-2, Decimators 77-0, 77-1, ..., 77-N-1, Analysis filter bank 78, Amplifiers 79-0, 79-1, ..., 79-N-1, A / D converters 80-0, 80- 1, ..., 80-N-1, and a decorrelator 81 having a memory 81a.
[0040]
The signal received via the antenna 71 passes through the BPF 72 and is RF-amplified by the LNA 73 and converted into two low IF signals having a quadrature phase relationship by the mixer 74. The converted signals are sequentially delayed by delay units 76-0, 76-1,..., 76-N-2, and then reduced to 1 / N by decimators 77-0, 77-1,. Down-sampled and input to the analysis filter bank 78. Thereby, the conversion speed of the A / D converters 80-0, 80-1,..., 80-N-1 is set to 1 / N. The analysis filter bank 78 divides the input signal into a plurality of frequency components and outputs a signal of each subband. Each subband signal is amplified by amplifiers 79-0, 79-1,..., 79-N-1, and converted into digital signals by A / D converters 80-0, 80-1,. After the conversion, the decorrelator 81 removes the mutual interference component from the input signal.
[0041]
Here, it is assumed that the frequency component of the transmission signal has d = {d ₀ , d ₁ ,..., D _N−1 } ^T. At this time, the received signal is expressed as follows.
[0042]
[Expression 12]

Here, ΔT is the sampling interval, and N is the number of inputs to the analysis filter bank 78. When there is no error in the coefficients of the analysis filter bank 78, the k-th output is as follows.
[0043]
[Formula 13]

However, the amplification degree of the amplifiers 79-0, 79-1,..., 79-N-1 is set to 1 for simplification.
[0044]
In the expression (2), the influence of the coefficient error and noise of the analysis filter is ignored. Actually, the coefficient of the analysis filter includes an error, and the received signal includes noise.
[0045]
[Expression 14]

Here, _nk represents the error of the nth coefficient of the filter for extracting the kth band, and η represents thermal noise. When Expression (3) is displayed in a matrix,
[Expression 15]

Further, when the decimated signal passes through each A / D converter 80-0, 80-1, ..., 80-N-1, quantization noise is added. If this is ζ = {ζ ₀ , ζ ₁ ,..., Ζ _N−1 } ^T ,
[Expression 16]

In the present embodiment, the coefficients of the analysis filter bank 78 are estimated in advance using a DC offset, the design values are compared with the estimated values, the cross-correlation matrix between the respective frequency-divided signals is calculated, and the cross-correlation is calculated. An inverse matrix of the matrix is calculated and stored in the memory 81a. In the decorrelator 81, the inverse matrix of hatR is multiplied by the outputs of the A / D converters 80-0, 80-1, ..., 80-N-1. That is, [0048]
[Expression 17]

The output of the decorrelator includes each frequency signal component d _k and a signal in which thermal noise η and quantization noise ζ are affected by cross-correlation.
[0049]
Here, the error rate characteristic of a conventional receiver that does not use a decorrelator is calculated. The desired signal received from equation (10) and the interference components from other frequency bands are expressed as follows.
[0050]
[Expression 18]

Here, Re {a} represents the real part of the complex number a, and [A] _k means the k-th element of the N × 1 matrix A. Also, the variance of noise due to the effect of thermal noise is given by equation (8):
[Equation 19]

Where σ ² is the variance of the thermal noise. Therefore [0052]
[Expression 20]

Also, assuming that the absolute value of the maximum input value of the A / D converter is A _M , the resolution is Bq bits, and the first 1 bit is used for a positive / negative sign, the quantization noise ζ _k is [− (A _M / 2 ^Bq ). , (A _M / 2 ^Bq )].
[0053]
If d _k is assumed to be a BPSK (Binary Phase Shift Keying) signal, the conditional error rate is obtained as follows.
[0054]
[Expression 21]

Here, [A] k, k means the (k, k) -th element of the N × N matrix A.
[0055]
[Expression 22]

And variable conversion and the error function is expanded.
[Expression 23]

Here [0057]
[Expression 24]

Therefore, the error rate of the receiver with compensation is
[Expression 25]

Here, the characteristics of the decorrelator type error compensation method of the present embodiment are analyzed. The variance of the noise component is calculated in equation (10). About thermal noise [0059]
[Equation 26]

As for the quantization error, it is assumed that the variance of the quantization error generated in each A / D converter is equal to λ ^2.
[Expression 27]

Given in. The quantization noise has a uniform distribution, but is modeled as Gaussian noise by multiplying the coefficients of the decorrelator. Therefore, the SNR (SN ratio) in the kth frequency band is obtained as follows.
[0061]
[Expression 28]

Here, [A] kk means the (k, k) -th element of the N × N matrix A. Assuming that d _k is a signal subjected to BPSK modulation and the noise component is Gaussian, the BER is obtained as follows.
[0062]
[Expression 29]

Here, as a simulation, an average error rate of 10,000 times is calculated when an error of [−0.01, 0.01] generated according to a uniform distribution is added to each coefficient of the analysis filter bank 78.
[0063]
It is assumed that the desired signal is BPSK modulated and transmitted, and is received and then down-converted to IF = 3π / 8 [rad / sample]. The communication path is assumed to be a white Gaussian noise path. Assuming coherent reception, one bit of the given resolution of the A / D converter is used to represent a positive or negative sign, and A _M = 2hatd _k or A _M = 2s _k (d). It is assumed that the image frequency signal is a BPSK signal received at −3π / 8 [rad / sample]. It is assumed that the power of the image frequency signal is 60 [dB] larger than the desired signal.
[0064]
FIG. 9 is a diagram showing the relationship between BER and Eb / No in the present embodiment. Here, the resolution of the A / D converter is 10 [bit]. In the case of a conventional receiver that does not use a decorrelator, the BER characteristic shows a floor due to the coefficient error of the filter and hardly changes with an increase in Eb / No. This is because the image frequency signal interferes with the desired signal due to a coefficient error. On the other hand, when the decorrelator type compensation method is used, the BER characteristic shows almost the theoretical value of BPSK. This is because the decorrelator removes the interference of the image frequency signal.
[0065]
The present invention is not limited to the above embodiment.
[0066]
In the above-described embodiment, a known signal is generated using a local signal by grounding the mixer input via a resistor. However, a local oscillator that generates a known signal is separately provided and supplied. Also good.
[0067]
Further, instead of using a filter bank as the filter configuration, a wavelet may be used. That is, instead of dividing the band into a large number at a time, it may be divided hierarchically, for example, so as to be repeatedly divided into two parts. In this way, it is possible to switch the fineness of division by, for example, switching that short-circuits a predetermined hierarchy to be divided.
[0068]
【The invention's effect】
As described above, according to the present invention, the coefficient error of the band division filter that is an analog part can be compensated by digital signal processing , and in particular, interference caused between the frequency signals orthogonal to each other due to the error of the analog filter. Can also be removed . Thereby, the request | requirement regarding the conversion speed and resolution of an A / D converter can be eased, and it can respond to a broadband communication system. Furthermore, parallel signal processing is possible, and a low power consumption reception configuration can be realized.
[Brief description of the drawings]
FIG. 1 is a diagram showing a configuration of a low-IF receiver shown as a reference .
FIG. 2 is a diagram showing a frequency response of a real coefficient QMF filter bank;
FIG. 3 is a diagram showing a frequency response of a QMF filter bank when the frequency is shifted by π / 4.
FIG. 4 is a diagram showing a configuration of a low IF receiver shown as a reference in the case of two divisions.
FIG. 5 is a diagram illustrating a relationship between a frequency shift amount of a filter and a BER.
FIG. 6 is a diagram showing the relationship between BER and Eb / No.
FIG. 7 is a diagram illustrating a relationship between BER and the resolution of an A / D converter.
FIG. 8 is a diagram showing a configuration of a low IF receiver according to an embodiment of the present invention.
FIG. 9 is a diagram showing the relationship between BER and Eb / No in the present embodiment.
FIG. 10 is a diagram illustrating a configuration of a conventional direct conversion receiver.
FIG. 11 is a diagram illustrating a configuration of a conventional low-IF receiver.
[Explanation of symbols]
11, 31, 71 Antenna 17, 38, 79 Amplifier 18, 39, 77 Decimator 20, 41 Upsampler 21, 42 Band synthesis filter

Claims (5)

A band division analog filter that divides the low IF signal into N (N is an integer of 2 or more) band signals;
An A / D converter for analog / digital conversion of the output of the band division analog filter;
Known signal supply means for supplying a low IF known signal to the low IF stage;
When a known signal is supplied to the band division analog filter by the known signal supply means ,
hat d = hat Rd
here,
hat d: (N , 1) matrix with output signal amplitude in each band as a component
d: (N , 1) matrix with input signal amplitude in each band as component
A memory for storing hat R ⁻¹ , which is an inverse matrix of the (N 1 , N) matrix hat R ,
A decorrelator that reads the contents of the memory and multiplies the output of the A / D converter by hat R ⁻¹ ;
A low IF receiver comprising a digital demodulator for demodulating the decorrelator output.   2. The low IF receiver according to claim 1, wherein said known signal supply means uses a local oscillator for converting a received signal frequency to a low IF.   3. The low IF receiver according to claim 1, wherein the band division analog filter has a filter bank configuration.   3. The low IF receiver according to claim 1, wherein the band division analog filter has a wavelet configuration.   5. The low IF receiver according to claim 1, wherein the band division analog filter has a complex coefficient filter configuration.

Images