1 |
2 |
danv |
--------------------------------------------------------------------------------
|
2 |
|
|
--
|
3 |
3 |
danv |
-- Copyright 2020
|
4 |
|
|
-- ASTRON (Netherlands Institute for Radio Astronomy) <http://www.astron.nl/>
|
5 |
|
|
-- P.O.Box 2, 7990 AA Dwingeloo, The Netherlands
|
6 |
|
|
--
|
7 |
|
|
-- Licensed under the Apache License, Version 2.0 (the "License");
|
8 |
|
|
-- you may not use this file except in compliance with the License.
|
9 |
|
|
-- You may obtain a copy of the License at
|
10 |
|
|
--
|
11 |
|
|
-- http://www.apache.org/licenses/LICENSE-2.0
|
12 |
|
|
--
|
13 |
|
|
-- Unless required by applicable law or agreed to in writing, software
|
14 |
|
|
-- distributed under the License is distributed on an "AS IS" BASIS,
|
15 |
|
|
-- WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
16 |
|
|
-- See the License for the specific language governing permissions and
|
17 |
|
|
-- limitations under the License.
|
18 |
2 |
danv |
--
|
19 |
|
|
--------------------------------------------------------------------------------
|
20 |
|
|
--
|
21 |
|
|
-- Purpose: The fft_sepa_wide unit performs the separate function on the
|
22 |
|
|
-- output of a complex wideband fft in order to extract the spectrum
|
23 |
|
|
-- of the two real inputs A and B. Where A was fed to the real input
|
24 |
|
|
-- of the complext wfft and B was fed to the imaginary input.
|
25 |
|
|
--
|
26 |
|
|
--
|
27 |
|
|
-- Description: The incoming data is stored in a dual paged ram. For each output
|
28 |
|
|
-- of the complex wfft a unique dual paged ram is instantiated. Once
|
29 |
|
|
-- the first page is written, the unit will read the data from the
|
30 |
|
|
-- memory.
|
31 |
|
|
-- The read process reads the memories in such a way that pairs of
|
32 |
|
|
-- data are created that are required to generate the correct outputs.
|
33 |
|
|
-- The data pairs are offered to the ZIP units that serialize the pairs.
|
34 |
|
|
-- The serialized data is then offered to the separate units that outputs
|
35 |
|
|
-- the separated data in an interleaved stream: A, B, A, B etc (for both real and imaginary part)
|
36 |
|
|
-- The last stage contains pipeline stages that are required for allignment
|
37 |
|
|
-- and additional pipeling.
|
38 |
|
|
--
|
39 |
|
|
|
40 |
5 |
danv |
library ieee, common_pkg_lib, astron_counter_lib, common_components_lib, astron_ram_lib, astron_multiplexer_lib;
|
41 |
2 |
danv |
use IEEE.std_logic_1164.all;
|
42 |
|
|
use IEEE.numeric_std.all;
|
43 |
|
|
use common_pkg_lib.common_pkg.all;
|
44 |
|
|
use work.fft_pkg.all;
|
45 |
|
|
|
46 |
|
|
entity fft_sepa_wide is
|
47 |
|
|
generic (
|
48 |
|
|
g_fft : t_fft := c_fft -- generics for the FFT
|
49 |
|
|
);
|
50 |
|
|
port (
|
51 |
|
|
clk : in std_logic;
|
52 |
|
|
rst : in std_logic := '0';
|
53 |
|
|
in_re_arr : in t_fft_slv_arr(g_fft.wb_factor-1 downto 0);
|
54 |
|
|
in_im_arr : in t_fft_slv_arr(g_fft.wb_factor-1 downto 0);
|
55 |
|
|
in_val : in std_logic := '1';
|
56 |
|
|
out_re_arr : out t_fft_slv_arr(g_fft.wb_factor-1 downto 0);
|
57 |
|
|
out_im_arr : out t_fft_slv_arr(g_fft.wb_factor-1 downto 0);
|
58 |
|
|
out_val : out std_logic
|
59 |
|
|
);
|
60 |
|
|
|
61 |
|
|
end entity fft_sepa_wide;
|
62 |
|
|
|
63 |
|
|
architecture rtl of fft_sepa_wide is
|
64 |
|
|
|
65 |
|
|
constant c_pipeline_output : natural := 0; -- no need for extra pipeline output, because output is already registered
|
66 |
|
|
|
67 |
|
|
constant c_page_size : natural := g_fft.nof_points/g_fft.wb_factor; -- Size of the memories
|
68 |
|
|
constant c_nof_pages : natural := 2; -- The number of pages in each ram.
|
69 |
|
|
constant c_dat_w : natural := c_nof_complex*g_fft.stage_dat_w; -- Data width for the internal vectors where real and imag are combined.
|
70 |
|
|
constant c_adr_w : natural := ceil_log2(c_page_size); -- Address width of the rams
|
71 |
|
|
constant c_nof_streams : natural := 2; -- Number of inputstreams for the zip units
|
72 |
|
|
|
73 |
|
|
type t_dat_arr is array(integer range <> ) of std_logic_vector(c_dat_w-1 downto 0);
|
74 |
|
|
type t_rd_adr_arr is array(integer range <> ) of std_logic_vector(c_adr_w-1 downto 0);
|
75 |
|
|
type t_zip_in_matrix is array(integer range <> ) of t_slv_64_arr(1 downto 0); -- Every Zip unit has two inputs.
|
76 |
|
|
|
77 |
|
|
signal next_page : std_logic; -- Active high signal to force a page-swap in the memories
|
78 |
|
|
signal wr_en : std_logic; -- The write enable signal for the memories
|
79 |
|
|
signal wr_adr : std_logic_vector(c_adr_w-1 downto 0); -- The write address
|
80 |
|
|
signal wr_dat : t_dat_arr(g_fft.wb_factor-1 downto 0); -- Array of data to be written to memory
|
81 |
|
|
|
82 |
|
|
signal rd_dat_arr : t_dat_arr(g_fft.wb_factor-1 downto 0); -- Array of data that is read from memory
|
83 |
|
|
signal rd_adr_arr : t_rd_adr_arr(1 downto 0); -- There are two different read addresses.
|
84 |
|
|
|
85 |
|
|
signal zip_in_matrix : t_zip_in_matrix(g_fft.wb_factor-1 downto 0); -- Matrix that contains the inputs for zip units
|
86 |
|
|
signal zip_in_val : std_logic_vector(g_fft.wb_factor-1 downto 0); -- Vector that holds the data input valids for the zip units
|
87 |
|
|
signal zip_out_dat_arr : t_dat_arr(g_fft.wb_factor-1 downto 0); -- Array that holds the outputs of all zip units.
|
88 |
|
|
signal zip_out_val : std_logic_vector(g_fft.wb_factor-1 downto 0); -- Vector that holds the output valids of the zip units
|
89 |
|
|
|
90 |
|
|
signal sep_out_dat_arr : t_dat_arr(g_fft.wb_factor-1 downto 0); -- Array that holds the outputs of the separation blocks
|
91 |
|
|
signal sep_out_val_vec : std_logic_vector(g_fft.wb_factor-1 downto 0); -- Vector containing the datavalids from the separation blocks
|
92 |
|
|
signal out_dat_arr : t_dat_arr(g_fft.wb_factor-1 downto 0); -- Array that holds the ouput values, where real and imag are concatenated
|
93 |
|
|
|
94 |
|
|
type state_type is (s_idle, s_read);
|
95 |
|
|
type reg_type is record
|
96 |
|
|
switch : std_logic; -- Toggle register used for separate functionalilty
|
97 |
|
|
count_up : natural range 0 to c_page_size; -- An upwards counter for read addressing
|
98 |
|
|
count_down : natural range 0 to c_page_size; -- A downwards counter for read addressing
|
99 |
|
|
val_odd : std_logic; -- Register that drives the in_valid of the odd zip units
|
100 |
|
|
val_even : std_logic; -- Register that drives the in_valid of the even zip units
|
101 |
|
|
state : state_type; -- The state machine.
|
102 |
|
|
end record;
|
103 |
|
|
|
104 |
|
|
signal r, rin : reg_type;
|
105 |
|
|
|
106 |
|
|
begin
|
107 |
|
|
|
108 |
|
|
---------------------------------------------------------------
|
109 |
|
|
-- DUAL PAGED RAMS
|
110 |
|
|
---------------------------------------------------------------
|
111 |
|
|
-- Prepare the data for the dual paged memory. Real and imaginary part are concatenated into one vector.
|
112 |
|
|
gen_prep_write_data : for I in 0 to g_fft.wb_factor-1 generate
|
113 |
|
|
wr_dat(I) <= in_im_arr(I)(g_fft.stage_dat_w-1 downto 0) & in_re_arr(I)(g_fft.stage_dat_w-1 downto 0);
|
114 |
|
|
end generate;
|
115 |
|
|
|
116 |
|
|
-- Prepare the write control signals for the memories.
|
117 |
|
|
wr_en <= in_val;
|
118 |
|
|
next_page <= '1' when unsigned(wr_adr) = c_page_size-1 and wr_en='1' else '0';
|
119 |
|
|
|
120 |
|
|
-- Counter will generate the write address
|
121 |
5 |
danv |
u_wr_adr_cnt : entity astron_counter_lib.common_counter
|
122 |
2 |
danv |
generic map(
|
123 |
|
|
g_latency => 1,
|
124 |
|
|
g_init => 0,
|
125 |
|
|
g_width => c_adr_w
|
126 |
|
|
)
|
127 |
|
|
port map (
|
128 |
|
|
rst => rst,
|
129 |
|
|
clk => clk,
|
130 |
|
|
cnt_en => in_val,
|
131 |
|
|
count => wr_adr
|
132 |
|
|
);
|
133 |
|
|
|
134 |
|
|
-- Instantiation of the rams.
|
135 |
|
|
gen_dual_paged_rams : for I in g_fft.wb_factor - 1 downto 0 generate
|
136 |
5 |
danv |
u_buff : entity astron_ram_lib.common_paged_ram_r_w
|
137 |
2 |
danv |
generic map (
|
138 |
|
|
g_str => "use_adr",
|
139 |
|
|
g_data_w => c_dat_w,
|
140 |
|
|
g_nof_pages => c_nof_pages,
|
141 |
|
|
g_page_sz => c_page_size,
|
142 |
|
|
g_wr_start_page => 0,
|
143 |
|
|
g_rd_start_page => 1,
|
144 |
|
|
g_rd_latency => 1
|
145 |
|
|
)
|
146 |
|
|
port map (
|
147 |
|
|
rst => rst,
|
148 |
|
|
clk => clk,
|
149 |
|
|
wr_next_page => next_page,
|
150 |
|
|
wr_adr => wr_adr,
|
151 |
|
|
wr_en => wr_en,
|
152 |
|
|
wr_dat => wr_dat(I),
|
153 |
|
|
rd_next_page => next_page,
|
154 |
|
|
rd_adr => rd_adr_arr(I/(g_fft.wb_factor/2)),
|
155 |
|
|
rd_en => '1',
|
156 |
|
|
rd_dat => rd_dat_arr(I),
|
157 |
|
|
rd_val => open
|
158 |
|
|
);
|
159 |
|
|
end generate;
|
160 |
|
|
|
161 |
|
|
-- Compose the read-addresses for the memories.
|
162 |
|
|
-- The first address toggles between the value of count_up and the value of count_up + offset.
|
163 |
|
|
-- The second address toggles between the value of count_down and the value of count_down + offset.
|
164 |
|
|
-- Note that the RESIZE_UVEC function generates the modulo(N) addressing.(The MSB is thrown away).
|
165 |
|
|
rd_adr_arr(0) <= RESIZE_UVEC(TO_UVEC(r.count_up, c_adr_w+1), c_adr_w) when r.switch = '0' else RESIZE_UVEC(TO_UVEC(r.count_up + c_page_size/2, c_adr_w+1), c_adr_w);
|
166 |
|
|
rd_adr_arr(1) <= RESIZE_UVEC(TO_UVEC(r.count_down, c_adr_w+1), c_adr_w) when r.switch = '0' else RESIZE_UVEC(TO_UVEC(r.count_down + c_page_size/2, c_adr_w+1), c_adr_w);
|
167 |
|
|
|
168 |
|
|
---------------------------------------------------------------
|
169 |
|
|
-- ZIP UNITS AND SEPARATORS
|
170 |
|
|
---------------------------------------------------------------
|
171 |
|
|
-- Compose the input matrix for the zip units. Each zip unit receives the
|
172 |
|
|
-- data of two different memories at the same time in order to allign the data
|
173 |
|
|
-- properly (in serial) for the separation units. Every zip unit receives data
|
174 |
|
|
-- once every two clock cylces.
|
175 |
|
|
gen_compose_zip_matrix : for I in g_fft.wb_factor/2 - 1 downto 0 generate
|
176 |
|
|
zip_in_matrix(2*I)(0) (c_dat_w-1 downto 0) <= rd_dat_arr(I);
|
177 |
|
|
zip_in_matrix(2*I)(1) (c_dat_w-1 downto 0) <= rd_dat_arr((g_fft.wb_factor-I) rem g_fft.wb_factor) when r.count_up = 0 else rd_dat_arr(g_fft.wb_factor-1-I);
|
178 |
|
|
zip_in_matrix(2*I+1)(0)(c_dat_w-1 downto 0) <= rd_dat_arr(I);
|
179 |
|
|
zip_in_matrix(2*I+1)(1)(c_dat_w-1 downto 0) <= rd_dat_arr(g_fft.wb_factor-1-I);
|
180 |
|
|
zip_in_val(2*I) <= r.val_even;
|
181 |
|
|
zip_in_val(2*I+1) <= r.val_odd;
|
182 |
|
|
end generate;
|
183 |
|
|
|
184 |
|
|
-- The instantiation of the zip units and the separation units.
|
185 |
|
|
-- The output of the zip units is connected to the input of the
|
186 |
|
|
-- adjacent separate unit.
|
187 |
|
|
gen_separators : for I in g_fft.wb_factor - 1 downto 0 generate
|
188 |
5 |
danv |
u_zipper : entity astron_multiplexer_lib.common_zip
|
189 |
2 |
danv |
generic map (
|
190 |
|
|
g_nof_streams => c_nof_streams,
|
191 |
|
|
g_dat_w => c_dat_w
|
192 |
|
|
)
|
193 |
|
|
port map (
|
194 |
|
|
rst => rst,
|
195 |
|
|
clk => clk,
|
196 |
|
|
in_val => zip_in_val(I),
|
197 |
|
|
in_dat_arr => zip_in_matrix(I),
|
198 |
|
|
out_val => zip_out_val(I),
|
199 |
|
|
out_dat => zip_out_dat_arr(I)
|
200 |
|
|
);
|
201 |
|
|
|
202 |
|
|
u_separate : entity work.fft_sepa
|
203 |
|
|
port map (
|
204 |
|
|
clk => clk,
|
205 |
|
|
rst => rst,
|
206 |
|
|
in_dat => zip_out_dat_arr(I),
|
207 |
|
|
in_val => zip_out_val(I),
|
208 |
|
|
out_dat => sep_out_dat_arr(I),
|
209 |
|
|
out_val => sep_out_val_vec(I)
|
210 |
|
|
);
|
211 |
|
|
end generate;
|
212 |
|
|
|
213 |
|
|
---------------------------------------------------------------
|
214 |
|
|
-- READ MEMORIES PROCESS
|
215 |
|
|
---------------------------------------------------------------
|
216 |
|
|
-- This process creates the read addresses for the dual page memories and
|
217 |
|
|
-- the fellow toggle signals. It also controls the starting and stopping
|
218 |
|
|
-- of the data stream.
|
219 |
|
|
comb : process(r, rst, next_page)
|
220 |
|
|
variable v : reg_type;
|
221 |
|
|
begin
|
222 |
|
|
|
223 |
|
|
v := r;
|
224 |
|
|
|
225 |
|
|
case r.state is
|
226 |
|
|
when s_idle =>
|
227 |
|
|
v.switch := '0';
|
228 |
|
|
v.val_odd := '0';
|
229 |
|
|
v.val_even := '0';
|
230 |
|
|
v.count_up := 0;
|
231 |
|
|
v.count_down := c_page_size;
|
232 |
|
|
if(next_page = '1') then -- Check if next page is asserted, meaning first page is written)
|
233 |
|
|
v.state := s_read;
|
234 |
|
|
end if;
|
235 |
|
|
|
236 |
|
|
when s_read =>
|
237 |
|
|
if(r.switch = '0') then -- Toggle the switch register from 0 to 1
|
238 |
|
|
v.switch := '1';
|
239 |
|
|
end if;
|
240 |
|
|
|
241 |
|
|
if(r.switch = '1') then -- Toggle the switch register from 1 to 0
|
242 |
|
|
v.switch := '0';
|
243 |
|
|
v.count_up := r.count_up + 1; -- Increment the upwards counter
|
244 |
|
|
v.count_down := r.count_down - 1; -- Decrease the downwards counter
|
245 |
|
|
end if;
|
246 |
|
|
|
247 |
|
|
if(next_page = '1') then -- Both counters are reset on page turn.
|
248 |
|
|
v.count_up := 0;
|
249 |
|
|
v.count_down := c_page_size;
|
250 |
|
|
elsif(v.count_up = c_page_size/2) then -- Pagereading is done, but there is not yet new data available (Note that the value of variable v is checked here)
|
251 |
|
|
v.state := s_idle; -- then go back to idle.
|
252 |
|
|
end if;
|
253 |
|
|
|
254 |
|
|
v.val_odd := r.switch; -- Assignment of the odd and even markers
|
255 |
|
|
v.val_even := not(r.switch);
|
256 |
|
|
|
257 |
|
|
when others =>
|
258 |
|
|
v.state := s_idle;
|
259 |
|
|
|
260 |
|
|
end case;
|
261 |
|
|
|
262 |
|
|
if(rst = '1') then
|
263 |
|
|
v.switch := '0';
|
264 |
|
|
v.count_up := 0;
|
265 |
|
|
v.count_down := 0;
|
266 |
|
|
v.val_odd := '0';
|
267 |
|
|
v.val_even := '0';
|
268 |
|
|
v.state := s_idle;
|
269 |
|
|
end if;
|
270 |
|
|
|
271 |
|
|
rin <= v;
|
272 |
|
|
|
273 |
|
|
end process comb;
|
274 |
|
|
|
275 |
|
|
regs : process(clk)
|
276 |
|
|
begin
|
277 |
|
|
if rising_edge(clk) then
|
278 |
|
|
r <= rin;
|
279 |
|
|
end if;
|
280 |
|
|
end process;
|
281 |
|
|
|
282 |
|
|
---------------------------------------------------------------
|
283 |
|
|
-- OUTPUT STAGE: ALIGNMENT AND PIPELINE STAGES
|
284 |
|
|
---------------------------------------------------------------
|
285 |
|
|
gen_align_and_pipeline_stages : for I in g_fft.wb_factor/2 - 1 downto 0 generate
|
286 |
|
|
u_output_pipeline_align : entity common_components_lib.common_pipeline
|
287 |
|
|
generic map (
|
288 |
|
|
g_pipeline => c_pipeline_output + 1, -- Pipeline + one stage for allignment
|
289 |
|
|
g_in_dat_w => c_dat_w,
|
290 |
|
|
g_out_dat_w => c_dat_w
|
291 |
|
|
)
|
292 |
|
|
port map (
|
293 |
|
|
clk => clk,
|
294 |
|
|
in_dat => sep_out_dat_arr(2*I),
|
295 |
|
|
out_dat => out_dat_arr(2*I)
|
296 |
|
|
);
|
297 |
|
|
|
298 |
|
|
u_output_pipeline : entity common_components_lib.common_pipeline
|
299 |
|
|
generic map (
|
300 |
|
|
g_pipeline => c_pipeline_output, -- Only pipeline stage
|
301 |
|
|
g_in_dat_w => c_dat_w,
|
302 |
|
|
g_out_dat_w => c_dat_w
|
303 |
|
|
)
|
304 |
|
|
port map (
|
305 |
|
|
clk => clk,
|
306 |
|
|
in_dat => sep_out_dat_arr(2*I+1),
|
307 |
|
|
out_dat => out_dat_arr(2*I+1)
|
308 |
|
|
);
|
309 |
|
|
end generate;
|
310 |
|
|
|
311 |
|
|
u_out_val_pipeline : entity common_components_lib.common_pipeline_sl
|
312 |
|
|
generic map (
|
313 |
|
|
g_pipeline => c_pipeline_output
|
314 |
|
|
)
|
315 |
|
|
port map (
|
316 |
|
|
clk => clk,
|
317 |
|
|
in_dat => sep_out_val_vec(1),
|
318 |
|
|
out_dat => out_val
|
319 |
|
|
);
|
320 |
|
|
|
321 |
|
|
-- Split the concatenated array into a real and imaginary array for the output
|
322 |
|
|
gen_output_arrays : for I in g_fft.wb_factor-1 downto 0 generate
|
323 |
|
|
out_re_arr(I) <= resize_fft_svec(out_dat_arr(I)( g_fft.stage_dat_w-1 downto 0));
|
324 |
|
|
out_im_arr(I) <= resize_fft_svec(out_dat_arr(I)(c_nof_complex*g_fft.stage_dat_w-1 downto g_fft.stage_dat_w));
|
325 |
|
|
end generate;
|
326 |
|
|
|
327 |
|
|
end rtl;
|
328 |
|
|
|
329 |
|
|
|